diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Cache/b7/9606fb3afea5bd1609ed40b622142f1c98125abcfe89a76a661b0e8e343910 b/.jekyll-cache/Jekyll/Cache/Jekyll--Cache/b7/9606fb3afea5bd1609ed40b622142f1c98125abcfe89a76a661b0e8e343910 index 20c6275..572369c 100644 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Cache/b7/9606fb3afea5bd1609ed40b622142f1c98125abcfe89a76a661b0e8e343910 +++ b/.jekyll-cache/Jekyll/Cache/Jekyll--Cache/b7/9606fb3afea5bd1609ed40b622142f1c98125abcfe89a76a661b0e8e343910 @@ -1 +1 @@ -I"ž{"source"=>"/Users/wanxinjin/Public/ASU Dropbox/Wanxin Jin/lab/website/asu-iris.github.io/asu-iris.github.io", "destination"=>"/Users/wanxinjin/Public/ASU Dropbox/Wanxin Jin/lab/website/asu-iris.github.io/asu-iris.github.io/_site", "collections_dir"=>"", "cache_dir"=>".jekyll-cache", "plugins_dir"=>"_plugins", "layouts_dir"=>"_layouts", "data_dir"=>"_data", "includes_dir"=>"_includes", "collections"=>{"posts"=>{"output"=>true, "permalink"=>"/:categories/:year/:month/:day/:title:output_ext"}}, "safe"=>false, "include"=>["_pages"], "exclude"=>["bin", "Gemfile", "Gemfile.lock", "vendor", ".sass-cache", ".jekyll-cache", "gemfiles", "node_modules", "vendor/bundle/", "vendor/cache/", "vendor/gems/", "vendor/ruby/"], "keep_files"=>["CNAME", ".nojekyll", ".git"], "encoding"=>"utf-8", "markdown_ext"=>"markdown,mkdown,mkdn,mkd,md", "strict_front_matter"=>false, "show_drafts"=>nil, "limit_posts"=>0, "future"=>false, "unpublished"=>false, "whitelist"=>[], "plugins"=>[], "markdown"=>"kramdown", "highlighter"=>"rouge", "lsi"=>false, "excerpt_separator"=>"\n\n", "incremental"=>false, "detach"=>false, "port"=>"4000", "host"=>"127.0.0.1", "baseurl"=>nil, "show_dir_listing"=>false, "permalink"=>"date", "paginate_path"=>"/page:num", "timezone"=>nil, "quiet"=>false, "verbose"=>false, "defaults"=>[], "liquid"=>{"error_mode"=>"warn", "strict_filters"=>false, "strict_variables"=>false}, "kramdown"=>{"auto_ids"=>true, "toc_levels"=>[1, 2, 3, 4, 5, 6], "entity_output"=>"as_char", "smart_quotes"=>"lsquo,rsquo,ldquo,rdquo", "input"=>"GFM", "hard_wrap"=>false, "guess_lang"=>true, "footnote_nr"=>1, "show_warnings"=>false, "syntax_highlighter_opts"=>{:css_class=>"highlight", :span=>{"line_numbers"=>false}, :block=>{"line_numbers"=>false, "start_line"=>1}, :default_lang=>"plaintext", :guess_lang=>true}, "syntax_highlighter"=>"rouge", "coderay"=>{}}, "title"=>"blank", "first_name"=>nil, "middle_name"=>nil, "last_name"=>"IRIS Lab", "email"=>"wanxinjin@gmail.com", "description"=>#, "footer_text"=>nil, "icon"=>nil, "url"=>"http://localhost:4000", "last_updated"=>9.2012, "impressum_path"=>nil, "navbar_fixed"=>true, "footer_fixed"=>true, "max_width"=>"900px", "serve_og_meta"=>false, "og_image"=>nil, "github_username"=>nil, "gitlab_username"=>nil, "twitter_username"=>nil, "linkedin_username"=>nil, "scholar_userid"=>nil, "orcid_id"=>nil, "medium_username"=>nil, "quora_username"=>nil, "publons_id"=>nil, "research_gate_profile"=>nil, "blogger_url"=>nil, "work_url"=>nil, "keybase_username"=>nil, "wikidata_id"=>nil, "dblp_url"=>nil, "stackoverflow_id"=>nil, "contact_note"=>nil, "google_analytics"=>"UA-XXXXXXXXX", "panelbear_analytics"=>"XXXXXXXXX", "highlight_theme"=>"github", "github"=>["metadata"], "enable_google_analytics"=>false, "enable_panelbear_analytics"=>false, "enable_mansory"=>true, "enable_math"=>true, "enable_tooltips"=>false, "enable_darkmode"=>false, "enable_navbar_social"=>false, "enable_project_categories"=>false, "enable_medium_zoom"=>false, "academicons"=>{"version"=>"1.9.0", "integrity"=>"sha512-W4yqoT1+8NLkinBLBZko+dFB2ZbHsYLDdr50VElllRcNt2Q4/GSs6u71UHKxB7S6JEMCp5Ve4xjh3eGQl/HRvg=="}, "bootstrap"=>{"version"=>"4.5.2", "integrity"=>{"css"=>"sha512-MoRNloxbStBcD8z3M/2BmnT+rg4IsMxPkXaGh2zD6LGNNFE80W3onsAhRcMAMrSoyWL9xD7Ert0men7vR8LUZg==", "js"=>"sha512-M5KW3ztuIICmVIhjSqXe01oV2bpe248gOxqmlcYrEzAvws7Pw3z6BK0iGbrwvdrUQUhi3eXgtxp5I8PDo9YfjQ=="}}, "fontawesome"=>{"version"=>"5.14.0", "integrity"=>"sha512-1PKOgIY59xJ8Co8+NE6FZ+LOAZKjy+KY8iq0G4B3CyeY6wYHN3yt9PW0XpSriVlkMXe40PTKnXrLnZ9+fkDaog=="}, "jquery"=>{"version"=>"3.5.1", "integrity"=>"sha512-bLT0Qm9VnAYZDflyKcBaQ2gg0hSYNQrJ8RilYldYQ1FxQYoCLtUjuuRuZo+fjqhx/qtq/1itJ0C2ejDxltZVFg=="}, "mathjax"=>{"version"=>"3.2.0"}, "mansory"=>{"version"=>"4.2.2", "integrity"=>"sha256-Nn1q/fx0H7SNLZMQ5Hw5JLaTRZp0yILA/FRexe19VdI="}, "mdb"=>{"version"=>"4.19.1", "integrity"=>{"css"=>"sha512-RO38pBRxYH3SoOprtPTD86JFOclM51/XTIdEPh5j8sj4tp8jmQIx26twG52UaLi//hQldfrh7e51WzP9wuP32Q==", "js"=>"sha512-Mug9KHKmroQFMLm93zGrjhibM2z2Obg9l6qFG2qKjXEXkMp/VDkI4uju9m4QKPjWSwQ6O2qzZEnJDEeCw0Blcw=="}}, "popper"=>{"version"=>"2.4.4", "integrity"=>"sha512-eUQ9hGdLjBjY3F41CScH3UX+4JDSI9zXeroz7hJ+RteoCaY+GP/LDoM8AO+Pt+DRFw3nXqsjh9Zsts8hnYv8/A=="}, "medium_zoom"=>{"version"=>"1.0.6", "integrity"=>"sha256-EdPgYcPk/IIrw7FYeuJQexva49pVRZNmt3LculEr7zM="}, "livereload_port"=>35729, "serving"=>true, "watch"=>true, "scholar"=>{"style"=>"apa", "locale"=>"en", "sort_by"=>"none", "order"=>"ascending", "group_by"=>"none", "group_order"=>"ascending", "bibliography_group_tag"=>"h2,h3,h4,h5", "bibliography_list_tag"=>"ol", "bibliography_item_tag"=>"li", "bibliography_list_attributes"=>{}, "bibliography_item_attributes"=>{}, "source"=>"./_bibliography", "bibliography"=>"references.bib", "repository"=>nil, "repository_file_delimiter"=>".", "bibtex_options"=>{:strip=>false, :parse_months=>true}, "bibtex_filters"=>[:smallcaps, :superscript, :italics, :textit, :lowercase, :textregistered, :tiny, :latex], "raw_bibtex_filters"=>[], "bibtex_skip_fields"=>[:abstract, :month_numeric], "bibtex_quotes"=>["{", "}"], "replace_strings"=>true, "join_strings"=>true, "remove_duplicates"=>false, "details_dir"=>"bibliography", "details_layout"=>"bibtex.html", "details_link"=>"Details", "details_permalink"=>"/:details_dir/:key:extension", "bibliography_class"=>"bibliography", "bibliography_template"=>"{{reference}}", "reference_tagname"=>"span", "missing_reference"=>"(missing reference)", "details_link_class"=>"details", "query"=>"@*", "cite_class"=>"citation", "type_names"=>{"article"=>"Journal Articles", "book"=>"Books", "incollection"=>"Book Chapters", "inproceedings"=>"Conference Articles", "thesis"=>"Theses", "mastersthesis"=>"Master's Theses", "phdthesis"=>"PhD Theses", "manual"=>"Manuals", "techreport"=>"Technical Reports", "misc"=>"Miscellaneous", "unpublished"=>"Unpublished"}, "type_aliases"=>{"phdthesis"=>"thesis", "mastersthesis"=>"thesis"}, "type_order"=>[], "month_names"=>nil}}:ET \ No newline at end of file +I"Š{"source"=>"/Users/wjin34/ASU Dropbox/Wanxin Jin/lab/website/asu-iris.github.io/asu-iris.github.io", "destination"=>"/Users/wjin34/ASU Dropbox/Wanxin Jin/lab/website/asu-iris.github.io/asu-iris.github.io/_site", "collections_dir"=>"", "cache_dir"=>".jekyll-cache", "plugins_dir"=>"_plugins", "layouts_dir"=>"_layouts", "data_dir"=>"_data", "includes_dir"=>"_includes", "collections"=>{"posts"=>{"output"=>true, "permalink"=>"/:categories/:year/:month/:day/:title:output_ext"}}, "safe"=>false, "include"=>["_pages"], "exclude"=>["bin", "Gemfile", "Gemfile.lock", "vendor", ".sass-cache", ".jekyll-cache", "gemfiles", "node_modules", "vendor/bundle/", "vendor/cache/", "vendor/gems/", "vendor/ruby/"], "keep_files"=>["CNAME", ".nojekyll", ".git"], "encoding"=>"utf-8", "markdown_ext"=>"markdown,mkdown,mkdn,mkd,md", "strict_front_matter"=>false, "show_drafts"=>nil, "limit_posts"=>0, "future"=>false, "unpublished"=>false, "whitelist"=>[], "plugins"=>[], "markdown"=>"kramdown", "highlighter"=>"rouge", "lsi"=>false, "excerpt_separator"=>"\n\n", "incremental"=>false, "detach"=>false, "port"=>"4000", "host"=>"127.0.0.1", "baseurl"=>nil, "show_dir_listing"=>false, "permalink"=>"date", "paginate_path"=>"/page:num", "timezone"=>nil, "quiet"=>false, "verbose"=>false, "defaults"=>[], "liquid"=>{"error_mode"=>"warn", "strict_filters"=>false, "strict_variables"=>false}, "kramdown"=>{"auto_ids"=>true, "toc_levels"=>[1, 2, 3, 4, 5, 6], "entity_output"=>"as_char", "smart_quotes"=>"lsquo,rsquo,ldquo,rdquo", "input"=>"GFM", "hard_wrap"=>false, "guess_lang"=>true, "footnote_nr"=>1, "show_warnings"=>false, "syntax_highlighter_opts"=>{:css_class=>"highlight", :span=>{"line_numbers"=>false}, :block=>{"line_numbers"=>false, "start_line"=>1}, :default_lang=>"plaintext", :guess_lang=>true}, "syntax_highlighter"=>"rouge", "coderay"=>{}}, "title"=>"blank", "first_name"=>nil, "middle_name"=>nil, "last_name"=>"IRIS Lab", "email"=>"wanxinjin@gmail.com", "description"=>#, "footer_text"=>nil, "icon"=>nil, "url"=>"http://localhost:4000", "last_updated"=>9.2012, "impressum_path"=>nil, "navbar_fixed"=>true, "footer_fixed"=>true, "max_width"=>"900px", "serve_og_meta"=>false, "og_image"=>nil, "github_username"=>nil, "gitlab_username"=>nil, "twitter_username"=>nil, "linkedin_username"=>nil, "scholar_userid"=>nil, "orcid_id"=>nil, "medium_username"=>nil, "quora_username"=>nil, "publons_id"=>nil, "research_gate_profile"=>nil, "blogger_url"=>nil, "work_url"=>nil, "keybase_username"=>nil, "wikidata_id"=>nil, "dblp_url"=>nil, "stackoverflow_id"=>nil, "contact_note"=>nil, "google_analytics"=>"UA-XXXXXXXXX", "panelbear_analytics"=>"XXXXXXXXX", "highlight_theme"=>"github", "github"=>["metadata"], "enable_google_analytics"=>false, "enable_panelbear_analytics"=>false, "enable_mansory"=>true, "enable_math"=>true, "enable_tooltips"=>false, "enable_darkmode"=>false, "enable_navbar_social"=>false, "enable_project_categories"=>false, "enable_medium_zoom"=>false, "academicons"=>{"version"=>"1.9.0", "integrity"=>"sha512-W4yqoT1+8NLkinBLBZko+dFB2ZbHsYLDdr50VElllRcNt2Q4/GSs6u71UHKxB7S6JEMCp5Ve4xjh3eGQl/HRvg=="}, "bootstrap"=>{"version"=>"4.5.2", "integrity"=>{"css"=>"sha512-MoRNloxbStBcD8z3M/2BmnT+rg4IsMxPkXaGh2zD6LGNNFE80W3onsAhRcMAMrSoyWL9xD7Ert0men7vR8LUZg==", "js"=>"sha512-M5KW3ztuIICmVIhjSqXe01oV2bpe248gOxqmlcYrEzAvws7Pw3z6BK0iGbrwvdrUQUhi3eXgtxp5I8PDo9YfjQ=="}}, "fontawesome"=>{"version"=>"5.14.0", "integrity"=>"sha512-1PKOgIY59xJ8Co8+NE6FZ+LOAZKjy+KY8iq0G4B3CyeY6wYHN3yt9PW0XpSriVlkMXe40PTKnXrLnZ9+fkDaog=="}, "jquery"=>{"version"=>"3.5.1", "integrity"=>"sha512-bLT0Qm9VnAYZDflyKcBaQ2gg0hSYNQrJ8RilYldYQ1FxQYoCLtUjuuRuZo+fjqhx/qtq/1itJ0C2ejDxltZVFg=="}, "mathjax"=>{"version"=>"3.2.0"}, "mansory"=>{"version"=>"4.2.2", "integrity"=>"sha256-Nn1q/fx0H7SNLZMQ5Hw5JLaTRZp0yILA/FRexe19VdI="}, "mdb"=>{"version"=>"4.19.1", "integrity"=>{"css"=>"sha512-RO38pBRxYH3SoOprtPTD86JFOclM51/XTIdEPh5j8sj4tp8jmQIx26twG52UaLi//hQldfrh7e51WzP9wuP32Q==", "js"=>"sha512-Mug9KHKmroQFMLm93zGrjhibM2z2Obg9l6qFG2qKjXEXkMp/VDkI4uju9m4QKPjWSwQ6O2qzZEnJDEeCw0Blcw=="}}, "popper"=>{"version"=>"2.4.4", "integrity"=>"sha512-eUQ9hGdLjBjY3F41CScH3UX+4JDSI9zXeroz7hJ+RteoCaY+GP/LDoM8AO+Pt+DRFw3nXqsjh9Zsts8hnYv8/A=="}, "medium_zoom"=>{"version"=>"1.0.6", "integrity"=>"sha256-EdPgYcPk/IIrw7FYeuJQexva49pVRZNmt3LculEr7zM="}, "livereload_port"=>35729, "serving"=>true, "watch"=>true, "scholar"=>{"style"=>"apa", "locale"=>"en", "sort_by"=>"none", "order"=>"ascending", "group_by"=>"none", "group_order"=>"ascending", "bibliography_group_tag"=>"h2,h3,h4,h5", "bibliography_list_tag"=>"ol", "bibliography_item_tag"=>"li", "bibliography_list_attributes"=>{}, "bibliography_item_attributes"=>{}, "source"=>"./_bibliography", "bibliography"=>"references.bib", "repository"=>nil, "repository_file_delimiter"=>".", "bibtex_options"=>{:strip=>false, :parse_months=>true}, "bibtex_filters"=>[:smallcaps, :superscript, :italics, :textit, :lowercase, :textregistered, :tiny, :latex], "raw_bibtex_filters"=>[], "bibtex_skip_fields"=>[:abstract, :month_numeric], "bibtex_quotes"=>["{", "}"], "replace_strings"=>true, "join_strings"=>true, "remove_duplicates"=>false, "details_dir"=>"bibliography", "details_layout"=>"bibtex.html", "details_link"=>"Details", "details_permalink"=>"/:details_dir/:key:extension", "bibliography_class"=>"bibliography", "bibliography_template"=>"{{reference}}", "reference_tagname"=>"span", "missing_reference"=>"(missing reference)", "details_link_class"=>"details", "query"=>"@*", "cite_class"=>"citation", "type_names"=>{"article"=>"Journal Articles", "book"=>"Books", "incollection"=>"Book Chapters", "inproceedings"=>"Conference Articles", "thesis"=>"Theses", "mastersthesis"=>"Master's Theses", "phdthesis"=>"PhD Theses", "manual"=>"Manuals", "techreport"=>"Technical Reports", "misc"=>"Miscellaneous", "unpublished"=>"Unpublished"}, "type_aliases"=>{"phdthesis"=>"thesis", "mastersthesis"=>"thesis"}, "type_order"=>[], "month_names"=>nil}}:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/00/6db4f7bf4ae300f911613038f65908c54b4ccda2869ac7985ef272bf21c71b b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/00/6db4f7bf4ae300f911613038f65908c54b4ccda2869ac7985ef272bf21c71b deleted file mode 100644 index 907b44f..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/00/6db4f7bf4ae300f911613038f65908c54b4ccda2869ac7985ef272bf21c71b +++ /dev/null @@ -1,130 +0,0 @@ -I"=%

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
August 19, 2024
-
-

We just released a major update to our product that includes several new features and improvements. Check out the video below for a walkthrough.

-
- -
-
-
- -
-
July 9 2024
-
-

- Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
- Check a breaf introduction of the paper: -
-
- -
-
-
- - -
- -
- -
-
July 9 2024
-
-

- Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -
-
- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -

- -
-
-
- Check a breaf introduction of the paper: -
- -
- -
- -
- Check out the project website and arXiv preprint. - </div> -</div> -</div> - -



- - - - -
July 9 2024
-
-

- Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -
-
- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -

- -
-
-
- Check a breaf introduction of the paper: -
- -
- -
- -
- Check out the project website and arXiv preprint. -

-:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/01/911be80ece26ee5195d4aa4ef024b0cb5350053be6052638e7387868a7da64 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/01/911be80ece26ee5195d4aa4ef024b0cb5350053be6052638e7387868a7da64 deleted file mode 100644 index 917efc9..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/01/911be80ece26ee5195d4aa4ef024b0cb5350053be6052638e7387868a7da64 +++ /dev/null @@ -1,351 +0,0 @@ -I"ëE - -

- -

The IRIS lab focuses on three reserach directions: (1) human-robot alignment, (2) contact-rich manipulation, and (3) fundamental methods in robotics. Below are some recent publications in each set of research interest. -Please visit Publications page for a full list of publications.

- -


-
-
Human-robot alignment
-
- -We develop methods to empower a robot with the ability to efficiently understand and be understood by human users through a variety of physical interactions. We explore how robots can aptly respond to and collaborate meaningfully with users. - -
-
-
    -
  • Robot learning from general human interactions
    • -
  • Planning and control for human-robot systems
    • -
-
- - -
- -
-
Safe MPC Alignment with Human Directional Feedback
-
Zhixian Xie, Wenlong Zhang, Yi Ren, Zhaoran Wang, George. J. Pappas and Wanxin Jin
-
Submitted to IEEE Transactions on Robotics (T-RO), 2024
-
- Paper - Code - Video - Project -
-
-
-
- - - -
- -
-
Learning from Human Directional Corrections
-
Wanxin Jin, Todd D Murphey, Zehui Lu, and Shaoshuai Mou
-
IEEE Transactions on Robotics (T-RO), 2023
-
- Paper - Code - Video -
-
-
-
- - -
- -
-
Learning from Sparse Demonstrations
-
Wanxin Jin, Todd D Murphey, Dana Kulic, Neta Ezer, and Shaoshuai Mou
-
IEEE Transactions on Robotics (T-RO), 2023
-
- Paper - Code - Video -
-
-
-
- - - - -
- -
-
Inverse Optimal Control from Incomplete Trajectory Observations
-
Wanxin Jin, Dana Kulic, Shaoshuai Mou, and Sandra Hirche
-
International Journal of Robotics Research (IJRR), 40:848–865, 2021
-
- Paper - Code -
-
-
-
- - - - -
- -
-
Inverse Optimal Control for Multiphase cost functions
-
Wanxin Jin, Dana Kulic, Jonathan Lin, Shaoshuai Mou, and Sandra Hirche
-
IEEE Transactions on Robotics (T-RO), 35(6):1387–1398, 2019
-
- Paper - Code -
-
-
- - - -
-
- -


- -
-
Contact-rich manipulation
-
- - -We aim to leverage physical principles to develop efficient representations or models for robot's physical interaction with environments. We also focus on developing algorithms to enable robots efficiently and robustly manipulate their surroundings/objects through contact. - -

- -
    -
  • Learning, planning, and control for contact-rich manipulation
    • -
  • Computer vision and learnable geometry for dexterous manipulation
    • -
-
- - - -
- -
-
Task-Driven Hybrid Model Reduction for Dexterous Manipulation
-
Wanxin Jin and Michael Posa
-
IEEE Transactions on Robotics (T-RO), 2024
-
- Paper - Code - Video - Project -
-
-
-
- - - - -
- -
-
Adaptive Contact-Implicit Model Predictive Control with Online Residual Learning
-
Wei-Cheng Huang, Alp Aydinoglu, Wanxin Jin, Michael Posa
-
IEEE International Conference on Robotics and Automation (ICRA), 2024
-
- Paper - Video - Project -
-
-
-
- - - -
- -
-
Adaptive Barrier Smoothing for First-Order Policy Gradient with Contact Dynamics
-
Shenao Zhang, Wanxin Jin, Zhaoran Wang
-
International Conference on Machine Learning (ICML), 2023
-
- Paper -
-
-
-
- - - -
- -
-
Learning Linear Complementarity Systems
-
Wanxin Jin, Alp Aydinoglu, Mathew Halm, and Michael Posa
-
Learning for Dynamics and Control (L4DC), 2022
-
- Paper - Code -
-
-
- - - -
-
- -


- -
-
Fundamental methods in robotics
-
- -We focus on developing fundamental theories and algorithms for achieving efficient, safe, and robust robot intelligence. Our methods lie at the intersection of model-based (control and optimization) and data-driven approaches, harnessing the complementary benefits of both. - -

- -
    -
  • Optimal control, motion plannig, reinforcement learning
    • -
  • Differentiable optimization, inverse optimization
    • -
  • Hybrid system learning and control
    • -
-
- - - -
- -
-
Pontryagin Differentiable Programming: An End-to-End Learning and Control Framework
-
Wanxin Jin, Zhaoran Wang, Zhuoran Yang, and Shaoshuai Mou
-
Advances in Neural Information Processing Systems (NeurIPS), 2020
-
- Paper - Code - Video - Project -
-
-
-
- - -
- -
-
Safe Pontryagin Differentiable Programming
-
Wanxin Jin, Shaoshuai Mou, and George J. Pappas
-
Advances in Neural Information Processing Systems (NeurIPS), 2021
-
- Paper - Code - Video - Project -
-
-
-
- - -
- -
-
Robust Safe Learning and Control in Unknown Environments: An Uncertainty-Aware Control Barrier Function Approach
-
Jiacheng Li, Qingchen Liu, Wanxin Jin, Jiahu Qin, and Sandra Hirche
-
IEEE Robotics and Automation Letters (RA-L), 2023
-
- Paper - Video -
-
-
-
- - - - -
- -
-
Enforcing Hard Constraints with Soft Barriers: Safe-driven Reinforcement Learning in Unknown Stochastic Environments
-
Yixuan Wang, Simon Sinong Zhan, Ruochen Jiao, Zhilu Wang, Wanxin Jin, Zhuoran Yang, Zhaoran Wang, Chao Huang, Qi Zhu
-
International Conference on Machine Learning (ICML), 2023
-
- Paper -
-
-
-
- - -
- -
-
A Differential Dynamic Programming Framework for Inverse Reinforcement Learning
-
Kun Cao, Xinhang Xu, Wanxin Jin, Karl H. Johansson, and Lihua Xie
-
Submitted to IEEE Transactions on Robotics (T-RO), 2024
-
- Paper -
-
-
- - -
-
- - - - - - -

- - -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/02/c412ac41507948726089229c57dc0ab5903b6d29a0d080a7211fbf4ba3023c b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/02/c412ac41507948726089229c57dc0ab5903b6d29a0d080a7211fbf4ba3023c deleted file mode 100644 index b2e9ee9..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/02/c412ac41507948726089229c57dc0ab5903b6d29a0d080a7211fbf4ba3023c +++ /dev/null @@ -1,85 +0,0 @@ -I"Ü

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

- 🚀 We’re THRILLED to unveil our latest work: “Complementarity-Free Multi-Contact Modeling and Optimization” method, which consistently achieves state-of-the-art results across various complex tasks (check out the video below!) including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation—each with different objects. - - - - -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/03/41cd68488d547d021d827033379032e60c94f332ae6484617ecb85b2980395 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/03/41cd68488d547d021d827033379032e60c94f332ae6484617ecb85b2980395 deleted file mode 100644 index 9947e91..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/03/41cd68488d547d021d827033379032e60c94f332ae6484617ecb85b2980395 +++ /dev/null @@ -1,81 +0,0 @@ -I"^

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
- - - - - - - -
-
Aug 18 2024
-
- -
-
- - -
- - -
July 9 2024
-
- Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -
-
- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
-
- Check a breaf introduction of the paper: -
-
- -
- -
- Check out the project website and arXiv preprint. -
- -
-
-:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/03/4e20f0ae89bcfd51763f6a7a35e1cd828e701826f3f66dbebe00897a4910df b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/03/4e20f0ae89bcfd51763f6a7a35e1cd828e701826f3f66dbebe00897a4910df deleted file mode 100644 index 7d18b83..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/03/4e20f0ae89bcfd51763f6a7a35e1cd828e701826f3f66dbebe00897a4910df +++ /dev/null @@ -1,94 +0,0 @@ -I"ý

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

Anounce our latest work: "Complementarity-Free Multi-Contact Modeling and Optimization." Our method consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks (check out 🎥 below!), including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation, all with different objects. -

- Our method sets a new benchmark in dexterous manipulation: -
    -
  • 🎯 A 96.5% success rate across all tasks
    • -
  • ⚙️ Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • 🚀 Model predictive control running at 50-100 Hz for every task
    • -
-
- -
-
- Please check out -
-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/03/b186aa0369642ef31e8daabf2e3161caea035ed10b5da85fd9c04596600bc5 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/03/b186aa0369642ef31e8daabf2e3161caea035ed10b5da85fd9c04596600bc5 deleted file mode 100644 index 911621c..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/03/b186aa0369642ef31e8daabf2e3161caea035ed10b5da85fd9c04596600bc5 +++ /dev/null @@ -1,581 +0,0 @@ -I"ąc - - - - - - - - -
-All -Human-robot alignment -Contact-rich manipulation -Fundamental methods -
- - - -

- - - -
-

2024

-
-
- - -
- -
-
Complementarity-Free Multi-Contact Modeling and Optimization for Dexterous Manipulation
-
Wanxin Jin
-
arXiv preprint, 2024
-
- Paper - Code - Video -
-
-
-
- - -
- -
-
Safe MPC Alignment with Human Directional Feedback
-
Zhixian Xie, Wenlong Zhang, Yi Ren, Zhaoran Wang, George. J. Pappas and Wanxin Jin
-
Submitted to IEEE Transactions on Robotics (T-RO), 2024
-
- Paper - Code - Video - Project -
-
-
-
- - -
- -
-
A Differential Dynamic Programming Framework for Inverse Reinforcement Learning
-
Kun Cao, Xinhang Xu, Wanxin Jin, Karl H. Johansson, and Lihua Xie
-
Submitted to IEEE Transactions on Robotics (T-RO), 2024
-
- Paper -
-
-
-
- - -
- -
-
D3G: Learning Multi-robot Coordination from Demonstrations
-
Yizhi Zhou, Wanxin Jin, Xuan Wang
-
IEEE/RSJ International Conference on Intelligent Robots and Systems, 2024
-
- Paper - Code - Video -
-
-
-
- - -
- -
-
TacTID: High-performance Visuo-Tactile Sensor-based Terrain Identification for Legged Robots
-
Ziwu Song, Chenchang Li, Zhentan Quan, Shilong Mu, Xiaosa Li, Ziyi Zhao, Wanxin Jin, Chenye Wu, Wenbo Ding, Xiao-Ping Zhang
-
IEEE Sensors Journal, 2024
-
- Paper - -
-
-
-
- - -
- -
-
How Can LLM Guide RL? A Value-Based Approach
-
Shenao Zhang, Sirui Zheng, Shuqi Ke, Zhihan Liu, Wanxin Jin, Jianbo Yuan, Yingxiang Yang, Hongxia Yang, Zhaoran Wang
-
arXiv preprint, 2024
-
- Paper - Code - -
-
-
-
- - -
- -
-
Task-Driven Hybrid Model Reduction for Dexterous Manipulation
-
Wanxin Jin and Michael Posa
-
IEEE Transactions on Robotics (T-RO), 2024
-
- Paper - Code - Video - Project -
-
-
-
- - - -
- -
-
Adaptive Contact-Implicit Model Predictive Control with Online Residual Learning
-
Wei-Cheng Huang, Alp Aydinoglu, Wanxin Jin, Michael Posa
-
IEEE International Conference on Robotics and Automation (ICRA), 2024
-
- Paper - Video - Project -
-
-
-
- - - - - -

- - - -
-

2023

-
-
- - -
- -
-
Guaranteed Stabilization and Safety of Nonlinear Systems via Sliding Mode Control
-
Fan Ding, Jin Ke, Wanxin Jin, Jianping He, and Xiaoming Duan
-
IEEE Control Systems Letters, 2023
-
- Paper -
-
-
-
- - - - - - - - -
- -
-
Adaptive Barrier Smoothing for First-Order Policy Gradient with Contact Dynamics
-
Shenao Zhang, Wanxin Jin, Zhaoran Wang
-
International Conference on Machine Learning (ICML), 2023
-
- Paper -
-
-
-
- - - - -
- -
-
Enforcing Hard Constraints with Soft Barriers: Safe-driven Reinforcement Learning in Unknown Stochastic Environments
-
Yixuan Wang, Simon Sinong Zhan, Ruochen Jiao, Zhilu Wang, Wanxin Jin, Zhuoran Yang, Zhaoran Wang, Chao Huang, Qi Zhu
-
International Conference on Machine Learning (ICML), 2023
-
- Paper -
-
-
-
- - - - -
- -
-
Robust Safe Learning and Control in Unknown Environments: An Uncertainty-Aware Control Barrier Function Approach
-
Jiacheng Li, Qingchen Liu, Wanxin Jin, Jiahu Qin, and Sandra Hirche
-
IEEE Robotics and Automation Letters (RA-L), 2023
-
- Paper - Video -
-
-
-
- - - -
- -
-
D3G: Learning Multi-robot Coordination from Demonstrations
-
Xuan Wang, YiZhi Zhou, and Wanxin Jin
-
IEEE International Conference on Intelligent Robots and Systems (IROS), 2023.
-
- Paper - Code - Video -
-
-
-
- - - - - -
- -
-
Identifying Reaction-Aware Driving Styles of Stochastic Model Predictive Controlled Vehicles by Inverse Reinforcement Learning
-
Ni Dang, Tao Shi, Zengjie Zhang, Wanxin Jin, Marion Leibold, and Martin Buss
-
International Conference on Intelligent Transportation Systems (ITSC), 2023.
-
- Paper -
-
-
-
- - - - - - - -

- - -
-

2022

-
-
- - -
- -
-
Learning from Human Directional Corrections
-
Wanxin Jin, Todd D Murphey, Zehui Lu, and Shaoshuai Mou
-
IEEE Transactions on Robotics (T-RO), 2023
-
- Paper - Code - Video -
-
-
-
- - - - - -
- -
-
Learning from Sparse Demonstrations
-
Wanxin Jin, Todd D Murphey, Dana Kulic, Neta Ezer, and Shaoshuai Mou
-
IEEE Transactions on Robotics (T-RO), 2023
-
- Paper - Code - Video -
-
-
-
- - - - - - -
- -
-
Learning Linear Complementarity Systems
-
Wanxin Jin, Alp Aydinoglu, Mathew Halm, and Michael Posa
-
Learning for Dynamics and Control (L4DC), 2022
-
- Paper - Code -
-
-
-
- - - - - - - -
- -
-
Cooperative Tuning of Multi-Agent Optimal Control Systems
-
Zehui Lu, Wanxin Jin, Shaoshuai Mou, Brian D. O. Anderson
-
IEEE Conference on Decision and Control (CDC), 2022
-
- Paper -
-
-
-
- - - - - -

- - -
-

2021

-
-
- - -
- -
-
Inverse Optimal Control from Incomplete Trajectory Observations
-
Wanxin Jin, Dana Kulic, Shaoshuai Mou, and Sandra Hirche
-
International Journal of Robotics Research (IJRR), 40:848–865, 2021
-
- Paper - Code -
-
-
-
- - - -
- -
-
Safe Pontryagin Differentiable Programming
-
Wanxin Jin, Shaoshuai Mou, and George J. Pappas
-
Advances in Neural Information Processing Systems (NeurIPS), 2021
-
- Paper - Code - Video - Project -
-
-
-
- - - - - - -
- -
-
Distributed Inverse Optimal Control
-
Wanxin Jin and Shaoshuai Mou
-
Automatica, Volume 129, 2021
-
- Paper - Code -
-
-
-
- - - - - - - -
- -
-
Human-Automation Interaction for Assisting Novices to Emulate Experts by Inferring Task Objective Functions
-
Sooyung Byeon, Wanxin Jin, Dawei Sun, and Inseok Hwang
-
AIAA/IEEE 40th Digital Avionics Systems Conference (DASC) , 2021. Best Student Paper Finalist
-
- Paper -
-
-
-
- - - - - - - - - -

- - -
-

2020

-
-
- - - -
- -
-
Pontryagin Differentiable Programming: An End-to-End Learning and Control Framework
-
Wanxin Jin, Zhaoran Wang, Zhuoran Yang, and Shaoshuai Mou
-
Advances in Neural Information Processing Systems (NeurIPS), 2020
-
- Paper - Code - Video - Project -
-
-
-
- - - - - - -

- - -
-

2019

-
-
- - - - -
- -
-
Inverse Optimal Control for Multiphase cost functions
-
Wanxin Jin, Dana Kulic, Jonathan Lin, Shaoshuai Mou, and Sandra Hirche
-
IEEE Transactions on Robotics (T-RO), 35(6):1387–1398, 2019
-
- Paper - Code -
-
-
-
- - - - - - - - - - - - - - - - - -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/04/72a285b2908424f2ba0503ff36336e0c86482d8e0af2ba5621c8ab1a5e470e b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/04/72a285b2908424f2ba0503ff36336e0c86482d8e0af2ba5621c8ab1a5e470e deleted file mode 100644 index 359ebd3..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/04/72a285b2908424f2ba0503ff36336e0c86482d8e0af2ba5621c8ab1a5e470e +++ /dev/null @@ -1,93 +0,0 @@ -I"Č

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

Thrilled to anounce our latest work: "Complementarity-Free Multi-Contact Modeling and Optimization", consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks (check out 🎥 below!), including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation, all with different objects. -

- Our method sets a new benchmark in dexterous manipulation: -
    -
  • 🎯 A 96.5% success rate across all tasks
    • -
  • ⚙️ Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • 🚀 Model predictive control running at 50-100 Hz for every task
    • -
-
- -
-
- Please check out the paper and code (fun guaranteed) -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can be very human-effort efficient! Our method, Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning, or declaring the misspecification of the hypothesis space. -
- -
-
- Check out project website and arXiv preprint, and a breaf introduction below. -

-
- -
-
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/06/1b59aab7ebc18fba3d1ba4636c7d7bf01e56e5e8ac166976ab26de3113877d b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/06/1b59aab7ebc18fba3d1ba4636c7d7bf01e56e5e8ac166976ab26de3113877d deleted file mode 100644 index 455832d..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/06/1b59aab7ebc18fba3d1ba4636c7d7bf01e56e5e8ac166976ab26de3113877d +++ /dev/null @@ -1,369 +0,0 @@ -I"ŔI - -

- -

The IRIS lab focuses on three reserach directions: (1) human-robot alignment, (2) contact-rich manipulation, and (3) fundamental methods in robotics. Below are some recent publications in each set of research interest. -Please visit Publications page for a full list of publications.

- -


-
-
Human-robot alignment
-
- -We develop methods to empower a robot with the ability to efficiently understand and be understood by human users through a variety of physical interactions. We explore how robots can aptly respond to and collaborate meaningfully with users. - -
-
-
    -
  • Robot learning from general human interactions
    • -
  • Planning and control for human-robot systems
    • -
-
- - -
- -
-
Safe MPC Alignment with Human Directional Feedback
-
Zhixian Xie, Wenlong Zhang, Yi Ren, Zhaoran Wang, George. J. Pappas and Wanxin Jin
-
Submitted to IEEE Transactions on Robotics (T-RO), 2024
-
- Paper - Code - Video - Project -
-
-
-
- - - -
- -
-
Learning from Human Directional Corrections
-
Wanxin Jin, Todd D Murphey, Zehui Lu, and Shaoshuai Mou
-
IEEE Transactions on Robotics (T-RO), 2023
-
- Paper - Code - Video -
-
-
-
- - -
- -
-
Learning from Sparse Demonstrations
-
Wanxin Jin, Todd D Murphey, Dana Kulic, Neta Ezer, and Shaoshuai Mou
-
IEEE Transactions on Robotics (T-RO), 2023
-
- Paper - Code - Video -
-
-
-
- - - - -
- -
-
Inverse Optimal Control from Incomplete Trajectory Observations
-
Wanxin Jin, Dana Kulic, Shaoshuai Mou, and Sandra Hirche
-
International Journal of Robotics Research (IJRR), 40:848–865, 2021
-
- Paper - Code -
-
-
-
- - - - -
- -
-
Inverse Optimal Control for Multiphase cost functions
-
Wanxin Jin, Dana Kulic, Jonathan Lin, Shaoshuai Mou, and Sandra Hirche
-
IEEE Transactions on Robotics (T-RO), 35(6):1387–1398, 2019
-
- Paper - Code -
-
-
- - - -
-
- -


- -
-
Contact-rich manipulation
-
- - -We aim to leverage physical principles to develop efficient representations or models for robot's physical interaction with environments. We also focus on developing algorithms to enable robots efficiently and robustly manipulate their surroundings/objects through contact. - -

- -
    -
  • Learning, planning, and control for contact-rich manipulation
    • -
  • Computer vision and learnable geometry for dexterous manipulation
    • -
-
- - - -
- -
-
Complementarity-Free Multi-Contact Modeling and Optimization for Dexterous Manipulation
-
Wanxin Jin
-
arXiv preprint, 2024
-
- Paper - Code - Video -
-
-
-
- - - - -
- -
-
Task-Driven Hybrid Model Reduction for Dexterous Manipulation
-
Wanxin Jin and Michael Posa
-
IEEE Transactions on Robotics (T-RO), 2024
-
- Paper - Code - Video - Project -
-
-
-
- - - - -
- -
-
Adaptive Contact-Implicit Model Predictive Control with Online Residual Learning
-
Wei-Cheng Huang, Alp Aydinoglu, Wanxin Jin, Michael Posa
-
IEEE International Conference on Robotics and Automation (ICRA), 2024
-
- Paper - Video - Project -
-
-
-
- - - -
- -
-
Adaptive Barrier Smoothing for First-Order Policy Gradient with Contact Dynamics
-
Shenao Zhang, Wanxin Jin, Zhaoran Wang
-
International Conference on Machine Learning (ICML), 2023
-
- Paper -
-
-
-
- - - -
- -
-
Learning Linear Complementarity Systems
-
Wanxin Jin, Alp Aydinoglu, Mathew Halm, and Michael Posa
-
Learning for Dynamics and Control (L4DC), 2022
-
- Paper - Code -
-
-
- - - -
-
- -


- -
-
Fundamental methods in robotics
-
- -We focus on developing fundamental theories and algorithms for achieving efficient, safe, and robust robot intelligence. Our methods lie at the intersection of model-based (control and optimization) and data-driven approaches, harnessing the complementary benefits of both. - -

- -
    -
  • Optimal control, motion plannig, reinforcement learning
    • -
  • Differentiable optimization, inverse optimization
    • -
  • Hybrid system learning and control
    • -
-
- - - -
- -
-
Pontryagin Differentiable Programming: An End-to-End Learning and Control Framework
-
Wanxin Jin, Zhaoran Wang, Zhuoran Yang, and Shaoshuai Mou
-
Advances in Neural Information Processing Systems (NeurIPS), 2020
-
- Paper - Code - Video - Project -
-
-
-
- - -
- -
-
Safe Pontryagin Differentiable Programming
-
Wanxin Jin, Shaoshuai Mou, and George J. Pappas
-
Advances in Neural Information Processing Systems (NeurIPS), 2021
-
- Paper - Code - Video - Project -
-
-
-
- - -
- -
-
Robust Safe Learning and Control in Unknown Environments: An Uncertainty-Aware Control Barrier Function Approach
-
Jiacheng Li, Qingchen Liu, Wanxin Jin, Jiahu Qin, and Sandra Hirche
-
IEEE Robotics and Automation Letters (RA-L), 2023
-
- Paper - Video -
-
-
-
- - - - -
- -
-
Enforcing Hard Constraints with Soft Barriers: Safe-driven Reinforcement Learning in Unknown Stochastic Environments
-
Yixuan Wang, Simon Sinong Zhan, Ruochen Jiao, Zhilu Wang, Wanxin Jin, Zhuoran Yang, Zhaoran Wang, Chao Huang, Qi Zhu
-
International Conference on Machine Learning (ICML), 2023
-
- Paper -
-
-
-
- - -
- -
-
A Differential Dynamic Programming Framework for Inverse Reinforcement Learning
-
Kun Cao, Xinhang Xu, Wanxin Jin, Karl H. Johansson, and Lihua Xie
-
Submitted to IEEE Transactions on Robotics (T-RO), 2024
-
- Paper -
-
-
- - -
-
- - - - - - -

- - -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/07/93b31d5cb48ec14c1ce0a6436c4088a5478ec7b8d43324b93e05c8550dabc4 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/07/93b31d5cb48ec14c1ce0a6436c4088a5478ec7b8d43324b93e05c8550dabc4 deleted file mode 100644 index e5acc61..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/07/93b31d5cb48ec14c1ce0a6436c4088a5478ec7b8d43324b93e05c8550dabc4 +++ /dev/null @@ -1,78 +0,0 @@ -I"<

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/0f/93bd6a3de1be2093ea977d64515debb634c2fdd5210d8dbe13739babd2108f b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/0f/93bd6a3de1be2093ea977d64515debb634c2fdd5210d8dbe13739babd2108f deleted file mode 100644 index eedf760..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/0f/93bd6a3de1be2093ea977d64515debb634c2fdd5210d8dbe13739babd2108f +++ /dev/null @@ -1,77 +0,0 @@ -I"(

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
- - - - - - - -
-
Aug 18 2024
-
- - - - - - - -
July 9 2024
-
- Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -
-
- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
-
- Check a breaf introduction of the paper: -
-
- -
- -
- Check out the project website and arXiv preprint. -
-:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/10/99788f11a87d9a28182acaf40b695a3a246b1d394b489cd8fc5b375df55e32 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/10/99788f11a87d9a28182acaf40b695a3a246b1d394b489cd8fc5b375df55e32 deleted file mode 100644 index f1e5c6f..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/10/99788f11a87d9a28182acaf40b695a3a246b1d394b489cd8fc5b375df55e32 +++ /dev/null @@ -1,91 +0,0 @@ -I"É

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

Anounce our latest work: "Complementarity-Free Multi-Contact Modeling and Optimization." Our method consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks (check out 🎥 below!), including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation, all with different objects. -

- Our method sets a new benchmark in dexterous manipulation: -
    -
  • 🎯 A 96.5% success rate across all tasks
    • -
  • ⚙️ Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • 🚀 Model predictive control running at 50-100 Hz for every task
    • -
-
- -
-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/11/1bdf0d79a610df41f6803ca5eef821fe22a3b99695ae3a4ac0a3b44a70244b b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/11/1bdf0d79a610df41f6803ca5eef821fe22a3b99695ae3a4ac0a3b44a70244b deleted file mode 100644 index eb51ebd..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/11/1bdf0d79a610df41f6803ca5eef821fe22a3b99695ae3a4ac0a3b44a70244b +++ /dev/null @@ -1,93 +0,0 @@ -I"?

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

Thrilled to anounce our latest work: "Complementarity-Free Multi-Contact Modeling and Optimization", consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks (check out 🎥 below!), including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation, all with different objects. -

- Our method sets a new benchmark in dexterous manipulation: -
    -
  • 🎯 A 96.5% success rate across all tasks
    • -
  • ⚙️ Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • 🚀 Model predictive control running at 50-100 Hz for every task
    • -
-
- -
-
- Please check out the paper and code (fun guaranteed) -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can be very human-effort efficient! Our method, Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check out project website and arXiv preprint, and a breaf introduction below. -

-
- -
-
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/12/5150fed5a74067e1e9b128a055c71a57444499d63f97bc6f82e0c66eeadace b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/12/5150fed5a74067e1e9b128a055c71a57444499d63f97bc6f82e0c66eeadace deleted file mode 100644 index af4e7ae..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/12/5150fed5a74067e1e9b128a055c71a57444499d63f97bc6f82e0c66eeadace +++ /dev/null @@ -1,85 +0,0 @@ -I"ŕ

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

- 🚀 We’re THRILLED to unveil our latest work: “Complementarity-Free Multi-Contact Modeling and Optimization.” Our method consistently achieves state-of-the-art results across various complex tasks (check out the video below!) including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation—each with different objects. - - - - -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/14/9e59c4f5d45007cf2cddc8cfad34b1c4904f7a2dddd4b2b93eef10cdf5e6bf b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/14/9e59c4f5d45007cf2cddc8cfad34b1c4904f7a2dddd4b2b93eef10cdf5e6bf deleted file mode 100644 index b0990fe..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/14/9e59c4f5d45007cf2cddc8cfad34b1c4904f7a2dddd4b2b93eef10cdf5e6bf +++ /dev/null @@ -1,78 +0,0 @@ -I"L

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
- - - - - - - -
-
Aug 18 2024
- - - - - - - - -
July 9 2024
-
- Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -
-
- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
-
- Check a breaf introduction of the paper: -
- -
- -
- -
- Check out the project website and arXiv preprint. -
-:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/15/fee8ee955fc5ff37066a6e843cc220c7717fc92316eddef27ef2fdb4899518 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/15/fee8ee955fc5ff37066a6e843cc220c7717fc92316eddef27ef2fdb4899518 deleted file mode 100644 index e729ccb..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/15/fee8ee955fc5ff37066a6e843cc220c7717fc92316eddef27ef2fdb4899518 +++ /dev/null @@ -1,87 +0,0 @@ -I"

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

Anounce our latest work: "Complementarity-Free Multi-Contact Modeling and Optimization." Our method consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks (check out 🎥 below!), including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation, all with different objects. -

- - Our method sets a new benchmark in dexterous manipulation: -
    -
  • 🎯 A 96.5% success rate across all tasks
    • -
  • 🧠 Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • ⚡ Model predictive control running at 50-100 Hz for every task
    • -
-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1b/4303bf130a09a3dfe5185abc5e4eddab3e3a6eb6bde9cb505910fbbae57564 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1b/4303bf130a09a3dfe5185abc5e4eddab3e3a6eb6bde9cb505910fbbae57564 deleted file mode 100644 index 43dac31..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1b/4303bf130a09a3dfe5185abc5e4eddab3e3a6eb6bde9cb505910fbbae57564 +++ /dev/null @@ -1,86 +0,0 @@ -I"§

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

- 🚀 We’re THRILLED to unveil our latest work: “Complementarity-Free Multi-Contact Modeling and Optimization” method, which consistently achieves state-of-the-art results across various complex tasks (check out the video below!) including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation—each with different objects. - - Our sets a new benchmark in dexterous manipulation, including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation—each with different objects. - - - -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1b/45ab5bda95d9aa3dd436ba1da315e5435881f9eb14e7727f15c9eaa9bc444d b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1b/45ab5bda95d9aa3dd436ba1da315e5435881f9eb14e7727f15c9eaa9bc444d deleted file mode 100644 index c24e328..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1b/45ab5bda95d9aa3dd436ba1da315e5435881f9eb14e7727f15c9eaa9bc444d +++ /dev/null @@ -1,89 +0,0 @@ -I"p

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

- 🚀 We’re THRILLED to unveil our latest work: “Complementarity-Free Multi-Contact Modeling and Optimization” method, which consistently achieves state-of-the-art results across various complex tasks (check out the video below!) including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation—each with different objects. - - Our sets a new benchmark in dexterous manipulation, including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation—each with different objects. - - ✨ A 96.5% success rate across all tasks - 🎯 Precise manipulation: 11° reorientation error & 7.8 mm position error - ⚡ Model predictive control running at 50-100 Hz for every task - - -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1c/d51e2d97e85b3eb97ecfeeee02c504d6bfa18b116f9b1e3c90eccaf81010c4 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1c/d51e2d97e85b3eb97ecfeeee02c504d6bfa18b116f9b1e3c90eccaf81010c4 deleted file mode 100644 index 42634f9..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1c/d51e2d97e85b3eb97ecfeeee02c504d6bfa18b116f9b1e3c90eccaf81010c4 +++ /dev/null @@ -1,87 +0,0 @@ -I" 

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

- 🚀 THRILLED to unveil our latest work: “Complementarity-Free Multi-Contact Modeling and Optimization” method, which consistently achieves state-of-the-art results across various complex tasks (check out the video below!) including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation—each with different objects. - Our sets a new benchmark in dexterous manipulation, including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation—each with different objects. -

    -
  • ✨ A 96.5% success rate across all tasks
    • -
  • 🎯 Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • ⚡ Model predictive control running at 50-100 Hz for every task
    • -
-

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1d/393448867f19525d4c54c7e5c5f45c3e440b68113236c84737f5823288a1b6 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1d/393448867f19525d4c54c7e5c5f45c3e440b68113236c84737f5823288a1b6 deleted file mode 100644 index 27e3769..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1d/393448867f19525d4c54c7e5c5f45c3e440b68113236c84737f5823288a1b6 +++ /dev/null @@ -1,78 +0,0 @@ -I"A

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- 🤖 Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1d/a563dc4500819cda10761112e21cced388881b31491ce6f00804d9b2d002f1 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1d/a563dc4500819cda10761112e21cced388881b31491ce6f00804d9b2d002f1 deleted file mode 100644 index 14b446b..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1d/a563dc4500819cda10761112e21cced388881b31491ce6f00804d9b2d002f1 +++ /dev/null @@ -1,93 +0,0 @@ -I"H

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

Thrilled to anounce our latest work: "Complementarity-Free Multi-Contact Modeling and Optimization", consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks (check out 🎥 below!), including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation, all with different objects. -

- Our method sets a new benchmark in dexterous manipulation: -
    -
  • 🎯 A 96.5% success rate across all tasks
    • -
  • ⚙️ Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • 🚀 Model predictive control running at 50-100 Hz for every task
    • -
-
- -
-
- Please check out the paper and code (fun guaranteed) -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method, Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check out project website and arXiv preprint, and a breaf introduction below. -

-
- -
-
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1e/410054e0f6d162cb690da86d377ad32821feb607119f1daae783ddfc16798e b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1e/410054e0f6d162cb690da86d377ad32821feb607119f1daae783ddfc16798e deleted file mode 100644 index 913a390..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/1e/410054e0f6d162cb690da86d377ad32821feb607119f1daae783ddfc16798e +++ /dev/null @@ -1,94 +0,0 @@ -I"ę

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

Anounce our latest work: "Complementarity-Free Multi-Contact Modeling and Optimization." Our method consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks (check out 🎥 below!), including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation, all with different objects. -

- Our method sets a new benchmark in dexterous manipulation: -
    -
  • 🎯 A 96.5% success rate across all tasks
    • -
  • ⚙️ Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • 🚀 Model predictive control running at 50-100 Hz for every task
    • -
-
- -
-
- Please check out the paper, and Code (fun guaranteed) -
-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/20/8c4c470d90116d20b48f83b2bd3fde27a8d67eef8b49ae5fdefe3b75eefc40 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/20/8c4c470d90116d20b48f83b2bd3fde27a8d67eef8b49ae5fdefe3b75eefc40 deleted file mode 100644 index 1444ad3..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/20/8c4c470d90116d20b48f83b2bd3fde27a8d67eef8b49ae5fdefe3b75eefc40 +++ /dev/null @@ -1,96 +0,0 @@ -I"ŕ

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
- -
-
July 9 2024
-
-

- Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -
-
- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -

- -
-
-
- Check a breaf introduction of the paper: -
- -
- -
- -
- Check out the project website and arXiv preprint. - </div> -</div> -</div> - -



- - - - -
July 9 2024
-
-

- Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -
-
- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -

- -
-
-
- Check a breaf introduction of the paper: -
- -
- -
- -
- Check out the project website and arXiv preprint. -

-:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/21/85c1a6c9b6953c237d91d9b27b26aed2a6d599101353d371a25bc59cd5de99 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/21/85c1a6c9b6953c237d91d9b27b26aed2a6d599101353d371a25bc59cd5de99 deleted file mode 100644 index 910924a..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/21/85c1a6c9b6953c237d91d9b27b26aed2a6d599101353d371a25bc59cd5de99 +++ /dev/null @@ -1,581 +0,0 @@ -I"¸c - - - - - - - - -
-All -Human-robot alignment -Contact-rich manipulation -Fundamental methods -
- - - -

- - - -
-

2024

-
-
- - -
- -
-
Complementarity-Free Multi-Contact Modeling and Optimization for Dexterous Manipulation
-
Wanxin Jin
-
arXiv preprint, 2024
-
- Paper - Code - Video -
-
-
-
- - -
- -
-
Safe MPC Alignment with Human Directional Feedback
-
Zhixian Xie, Wenlong Zhang, Yi Ren, Zhaoran Wang, George. J. Pappas and Wanxin Jin
-
Submitted to IEEE Transactions on Robotics (T-RO), 2024
-
- Paper - Code - Video - Project -
-
-
-
- - -
- -
-
A Differential Dynamic Programming Framework for Inverse Reinforcement Learning
-
Kun Cao, Xinhang Xu, Wanxin Jin, Karl H. Johansson, and Lihua Xie
-
Submitted to IEEE Transactions on Robotics (T-RO), 2024
-
- Paper -
-
-
-
- - -
- -
-
D3G: Learning Multi-robot Coordination from Demonstrations
-
Yizhi Zhou, Wanxin Jin, Xuan Wang
-
IEEE/RSJ International Conference on Intelligent Robots and Systems, 2024
-
- Paper - Code - Video -
-
-
-
- - -
- -
-
TacTID: High-performance Visuo-Tactile Sensor-based Terrain Identification for Legged Robots
-
Ziwu Song, Chenchang Li, Zhentan Quan, Shilong Mu, Xiaosa Li, Ziyi Zhao, Wanxin Jin, Chenye Wu, Wenbo Ding, Xiao-Ping Zhang
-
IEEE Sensors Journal, 2024
-
- Paper - -
-
-
-
- - -
- -
-
How Can LLM Guide RL? A Value-Based Approach
-
Shenao Zhang, Sirui Zheng, Shuqi Ke, Zhihan Liu, Wanxin Jin, Jianbo Yuan, Yingxiang Yang, Hongxia Yang, Zhaoran Wang
-
arXiv preprint, 2024
-
- Paper - Code - -
-
-
-
- - -
- -
-
Task-Driven Hybrid Model Reduction for Dexterous Manipulation
-
Wanxin Jin and Michael Posa
-
IEEE Transactions on Robotics (T-RO), 2024
-
- Paper - Code - Video - Project -
-
-
-
- - - -
- -
-
Adaptive Contact-Implicit Model Predictive Control with Online Residual Learning
-
Wei-Cheng Huang, Alp Aydinoglu, Wanxin Jin, Michael Posa
-
IEEE International Conference on Robotics and Automation (ICRA), 2024
-
- Paper - Video - Project -
-
-
-
- - - - - -

- - - -
-

2023

-
-
- - -
- -
-
Guaranteed Stabilization and Safety of Nonlinear Systems via Sliding Mode Control
-
Fan Ding, Jin Ke, Wanxin Jin, Jianping He, and Xiaoming Duan
-
IEEE Control Systems Letters, 2023
-
- Paper -
-
-
-
- - - - - - - - -
- -
-
Adaptive Barrier Smoothing for First-Order Policy Gradient with Contact Dynamics
-
Shenao Zhang, Wanxin Jin, Zhaoran Wang
-
International Conference on Machine Learning (ICML), 2023
-
- Paper -
-
-
-
- - - - -
- -
-
Enforcing Hard Constraints with Soft Barriers: Safe-driven Reinforcement Learning in Unknown Stochastic Environments
-
Yixuan Wang, Simon Sinong Zhan, Ruochen Jiao, Zhilu Wang, Wanxin Jin, Zhuoran Yang, Zhaoran Wang, Chao Huang, Qi Zhu
-
International Conference on Machine Learning (ICML), 2023
-
- Paper -
-
-
-
- - - - -
- -
-
Robust Safe Learning and Control in Unknown Environments: An Uncertainty-Aware Control Barrier Function Approach
-
Jiacheng Li, Qingchen Liu, Wanxin Jin, Jiahu Qin, and Sandra Hirche
-
IEEE Robotics and Automation Letters (RA-L), 2023
-
- Paper - Video -
-
-
-
- - - -
- -
-
D3G: Learning Multi-robot Coordination from Demonstrations
-
Xuan Wang, YiZhi Zhou, and Wanxin Jin
-
IEEE International Conference on Intelligent Robots and Systems (IROS), 2023.
-
- Paper - Code - Video -
-
-
-
- - - - - -
- -
-
Identifying Reaction-Aware Driving Styles of Stochastic Model Predictive Controlled Vehicles by Inverse Reinforcement Learning
-
Ni Dang, Tao Shi, Zengjie Zhang, Wanxin Jin, Marion Leibold, and Martin Buss
-
International Conference on Intelligent Transportation Systems (ITSC), 2023.
-
- Paper -
-
-
-
- - - - - - - -

- - -
-

2022

-
-
- - -
- -
-
Learning from Human Directional Corrections
-
Wanxin Jin, Todd D Murphey, Zehui Lu, and Shaoshuai Mou
-
IEEE Transactions on Robotics (T-RO), 2023
-
- Paper - Code - Video -
-
-
-
- - - - - -
- -
-
Learning from Sparse Demonstrations
-
Wanxin Jin, Todd D Murphey, Dana Kulic, Neta Ezer, and Shaoshuai Mou
-
IEEE Transactions on Robotics (T-RO), 2023
-
- Paper - Code - Video -
-
-
-
- - - - - - -
- -
-
Learning Linear Complementarity Systems
-
Wanxin Jin, Alp Aydinoglu, Mathew Halm, and Michael Posa
-
Learning for Dynamics and Control (L4DC), 2022
-
- Paper - Code -
-
-
-
- - - - - - - -
- -
-
Cooperative Tuning of Multi-Agent Optimal Control Systems
-
Zehui Lu, Wanxin Jin, Shaoshuai Mou, Brian D. O. Anderson
-
IEEE Conference on Decision and Control (CDC), 2022
-
- Paper -
-
-
-
- - - - - -

- - -
-

2021

-
-
- - -
- -
-
Inverse Optimal Control from Incomplete Trajectory Observations
-
Wanxin Jin, Dana Kulic, Shaoshuai Mou, and Sandra Hirche
-
International Journal of Robotics Research (IJRR), 40:848–865, 2021
-
- Paper - Code -
-
-
-
- - - -
- -
-
Safe Pontryagin Differentiable Programming
-
Wanxin Jin, Shaoshuai Mou, and George J. Pappas
-
Advances in Neural Information Processing Systems (NeurIPS), 2021
-
- Paper - Code - Video - Project -
-
-
-
- - - - - - -
- -
-
Distributed Inverse Optimal Control
-
Wanxin Jin and Shaoshuai Mou
-
Automatica, Volume 129, 2021
-
- Paper - Code -
-
-
-
- - - - - - - -
- -
-
Human-Automation Interaction for Assisting Novices to Emulate Experts by Inferring Task Objective Functions
-
Sooyung Byeon, Wanxin Jin, Dawei Sun, and Inseok Hwang
-
AIAA/IEEE 40th Digital Avionics Systems Conference (DASC) , 2021. Best Student Paper Finalist
-
- Paper -
-
-
-
- - - - - - - - - -

- - -
-

2020

-
-
- - - -
- -
-
Pontryagin Differentiable Programming: An End-to-End Learning and Control Framework
-
Wanxin Jin, Zhaoran Wang, Zhuoran Yang, and Shaoshuai Mou
-
Advances in Neural Information Processing Systems (NeurIPS), 2020
-
- Paper - Code - Video - Project -
-
-
-
- - - - - - -

- - -
-

2019

-
-
- - - - -
- -
-
Inverse Optimal Control for Multiphase cost functions
-
Wanxin Jin, Dana Kulic, Jonathan Lin, Shaoshuai Mou, and Sandra Hirche
-
IEEE Transactions on Robotics (T-RO), 35(6):1387–1398, 2019
-
- Paper - Code -
-
-
-
- - - - - - - - - - - - - - - - - -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/21/8894ab23e88ba413a8ca42a9b9f2f83cf11b0ae88834b1446f6bc409030295 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/21/8894ab23e88ba413a8ca42a9b9f2f83cf11b0ae88834b1446f6bc409030295 deleted file mode 100644 index 3c9613b..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/21/8894ab23e88ba413a8ca42a9b9f2f83cf11b0ae88834b1446f6bc409030295 +++ /dev/null @@ -1,93 +0,0 @@ -I"č

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? YES! The key lies in our new "effective yet optimization-friendly multi-contact model". -

-

Thrilled to unveil our latest work: Complementarity-Free Multi-Contact Modeling and Optimization, which consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks, including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation, all with different objects. Check out the demo below! -

- Our method sets a new benchmark in dexterous manipulation: -
    -
  • 🎯 A 96.5% success rate across all tasks
    • -
  • ⚙️ High manipulation accuracy: 11° reorientation error & 7.8 mm position error
    • -
  • 🚀 Model predictive control running at 50-100 Hz for all tasks
    • -
-
- -
-
- Please check out the paper and code (fun guaranteed) -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check out the project website, preprint, and a breaf introduction vide below. -
-
- -
-
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/22/2907628dc05be440e4b55b7f3aea0073ccb2164af855a8c5ac4db051e471c0 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/22/2907628dc05be440e4b55b7f3aea0073ccb2164af855a8c5ac4db051e471c0 deleted file mode 100644 index bb54373..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/22/2907628dc05be440e4b55b7f3aea0073ccb2164af855a8c5ac4db051e471c0 +++ /dev/null @@ -1,93 +0,0 @@ -I"×

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

Thrilled to anounce our latest work: "Complementarity-Free Multi-Contact Modeling and Optimization", consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks (check out 🎥 below!), including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation, all with different objects. -

- Our method sets a new benchmark in dexterous manipulation: -
    -
  • 🎯 A 96.5% success rate across all tasks
    • -
  • ⚙️ Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • 🚀 Model predictive control running at 50-100 Hz for every task
    • -
-
- -
-
- Please check out the paper and code (fun guaranteed) -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check out project website and arXiv preprint breaf introduction of the paper: -
-
- -
-
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/24/23bb9170b37f92de73cbdd4d5f0328aabd2729c893ae129c1df1202e4585cc b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/24/23bb9170b37f92de73cbdd4d5f0328aabd2729c893ae129c1df1202e4585cc deleted file mode 100644 index 99e0bed..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/24/23bb9170b37f92de73cbdd4d5f0328aabd2729c893ae129c1df1202e4585cc +++ /dev/null @@ -1,87 +0,0 @@ -I"%

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

Anounce our latest work: "Complementarity-Free Multi-Contact Modeling and Optimization." Our method consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks (check out the video below!), including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation—each with different objects. - 🎥

- - Our method sets a new benchmark in dexterous manipulation: -
    -
  • ✨ A 96.5% success rate across all tasks
    • -
  • 🎯 Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • ⚡ Model predictive control running at 50-100 Hz for every task
    • -
-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/25/ad034cd33db0ffdc08cb3a8b82263fff5d00871df6b339d34a2eaace57c5ec b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/25/ad034cd33db0ffdc08cb3a8b82263fff5d00871df6b339d34a2eaace57c5ec deleted file mode 100644 index ea939c4..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/25/ad034cd33db0ffdc08cb3a8b82263fff5d00871df6b339d34a2eaace57c5ec +++ /dev/null @@ -1,89 +0,0 @@ -I"Ş

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

- 🚀 We’re THRILLED to unveil our latest work: “Complementarity-Free Multi-Contact Modeling and Optimization” method, which consistently achieves state-of-the-art results across various complex tasks (check out the video below!) including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation—each with different objects. - - Our sets a new benchmark in dexterous manipulation, including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation—each with different objects. -

    -
  • ✨ A 96.5% success rate across all tasks
    • -
  • 🎯 Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • ⚡ Model predictive control running at 50-100 Hz for every task
    • -
- -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/26/1350d20f6aea9fa97eea002fa0562739b440d8ef88dbca98b3c669b4f5f7ca b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/26/1350d20f6aea9fa97eea002fa0562739b440d8ef88dbca98b3c669b4f5f7ca deleted file mode 100644 index 9aebe48..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/26/1350d20f6aea9fa97eea002fa0562739b440d8ef88dbca98b3c669b4f5f7ca +++ /dev/null @@ -1,93 +0,0 @@ -I"Ň

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? YES! The key lies in our new "effective yet optimization-friendly multi-contact model". -

-

🔥 Thrilled to unveil our work: Complementarity-Free Multi-Contact Modeling and Optimization, which consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks, including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation, all with different objects. Check out the demo below! -

- Our method sets a new benchmark in dexterous manipulation: -
    -
  • 🎯 A 96.5% success rate across all tasks
    • -
  • ⚙️ High manipulation accuracy: 11° reorientation error & 7.8 mm position error
    • -
  • 🚀 Model predictive control running at 50-100 Hz for all tasks
    • -
-
- -
-
- Check out our preprint , and try out our code --- fun guaranteed! -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check out the project website, preprint, and a breaf introduction vide below. -
-
- -
-
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/29/80d7b085e700183ca788cd289c09ab9e27e158bc3a1b79640c6f6377819a0b b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/29/80d7b085e700183ca788cd289c09ab9e27e158bc3a1b79640c6f6377819a0b deleted file mode 100644 index 33ac612..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/29/80d7b085e700183ca788cd289c09ab9e27e158bc3a1b79640c6f6377819a0b +++ /dev/null @@ -1,93 +0,0 @@ -I"Ż

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? YES! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

Thrilled to anounce our latest work: "Complementarity-Free Multi-Contact Modeling and Optimization", consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks (check out 🎥 below!), including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation, all with different objects. -

- Our method sets a new benchmark in dexterous manipulation: -
    -
  • 🎯 A 96.5% success rate across all tasks
    • -
  • ⚙️ Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • 🚀 Model predictive control running at 50-100 Hz for every task
    • -
-
- -
-
- Please check out the paper and code (fun guaranteed) -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check out the project website, preprint, and a breaf introduction vide below. -
-
- -
-
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/2a/722315d1a4ebc2b7dddc9039033177dab328d266307449035dfe1b994d37a7 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/2a/722315d1a4ebc2b7dddc9039033177dab328d266307449035dfe1b994d37a7 deleted file mode 100644 index 74dffae..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/2a/722315d1a4ebc2b7dddc9039033177dab328d266307449035dfe1b994d37a7 +++ /dev/null @@ -1,87 +0,0 @@ -I";

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

🚀 We're THRILLED to unveil our latest work: "Complementarity-Free Multi-Contact Modeling and Optimization." Our method consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks (check out the video below!), including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation—each with different objects. - 🎥

- - Our method sets a new benchmark in dexterous manipulation: -
    -
  • ✨ A 96.5% success rate across all tasks
    • -
  • 🎯 Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • ⚡ Model predictive control running at 50-100 Hz for every task
    • -
-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/2b/574a3d6c4aaedae9ac5c8d5d0f22034ca561a491a4daf0c3b6561239ad2395 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/2b/574a3d6c4aaedae9ac5c8d5d0f22034ca561a491a4daf0c3b6561239ad2395 deleted file mode 100644 index 88507df..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/2b/574a3d6c4aaedae9ac5c8d5d0f22034ca561a491a4daf0c3b6561239ad2395 +++ /dev/null @@ -1,93 +0,0 @@ -I"Ý

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? YES! The key lies in our new "effective yet optimization-friendly multi-contact model". -

-

Thrilled to anounce our latest work: Complementarity-Free Multi-Contact Modeling and Optimization, consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks (check out 🎥 below!), including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation, all with different objects. -

- Our method sets a new benchmark in dexterous manipulation: -
    -
  • 🎯 A 96.5% success rate across all tasks
    • -
  • ⚙️ Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • 🚀 Model predictive control running at 50-100 Hz for every task
    • -
-
- -
-
- Please check out the paper and code (fun guaranteed) -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check out the project website, preprint, and a breaf introduction vide below. -
-
- -
-
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/30/ee0f6cda74addb2d17b1c4ae6964303255d711161dbd246851177884bf1b9c b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/30/ee0f6cda74addb2d17b1c4ae6964303255d711161dbd246851177884bf1b9c deleted file mode 100644 index 7b0d1a9..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/30/ee0f6cda74addb2d17b1c4ae6964303255d711161dbd246851177884bf1b9c +++ /dev/null @@ -1,87 +0,0 @@ -I"

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

Anounce our latest work: "Complementarity-Free Multi-Contact Modeling and Optimization." Our method consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks (check out 🎥below!), including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation—each with different objects. -

- - Our method sets a new benchmark in dexterous manipulation: -
    -
  • ✨ A 96.5% success rate across all tasks
    • -
  • 🎯 Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • ⚡ Model predictive control running at 50-100 Hz for every task
    • -
-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/31/0cac276d45519f15fb4e488266007089df9c625b81af7a6c027fd358c4d6a5 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/31/0cac276d45519f15fb4e488266007089df9c625b81af7a6c027fd358c4d6a5 deleted file mode 100644 index be7e657..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/31/0cac276d45519f15fb4e488266007089df9c625b81af7a6c027fd358c4d6a5 +++ /dev/null @@ -1,93 +0,0 @@ -I"É

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

Thrilled to anounce our latest work: "Complementarity-Free Multi-Contact Modeling and Optimization", consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks (check out 🎥 below!), including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation, all with different objects. -

- Our method sets a new benchmark in dexterous manipulation: -
    -
  • 🎯 A 96.5% success rate across all tasks
    • -
  • ⚙️ Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • 🚀 Model predictive control running at 50-100 Hz for every task
    • -
-
- -
-
- Please check out the paper and code (fun guaranteed) -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- Robots 🤖 may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can be very human-effort efficient! Our method, Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning, or declaring the misspecification of the hypothesis space. -
- -
-
- Check out project website and arXiv preprint, and a breaf introduction below. -

-
- -
-
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/32/0f396630771f9521ef6ae5666287f4f35f9b5dfb101a0713aa0a5f02e491b2 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/32/0f396630771f9521ef6ae5666287f4f35f9b5dfb101a0713aa0a5f02e491b2 deleted file mode 100644 index d5ffe3b..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/32/0f396630771f9521ef6ae5666287f4f35f9b5dfb101a0713aa0a5f02e491b2 +++ /dev/null @@ -1,96 +0,0 @@ -I"Ţ

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
- -
-
July 9 2024
-
-

- Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -
-
- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -

- -
-
-
- Check a breaf introduction of the paper: -
- -
- -
- -
- Check out the project website and arXiv preprint. - </div> -</div> -</div> - -



- - - - -
July 9 2024
-
-

- Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -
-
- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -

- -
-
-
- Check a breaf introduction of the paper: -
- -
- -
- -
- Check out the project website and arXiv preprint. -

-:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/33/b703df2942bd2acbfaeb55cd5d4fca107ab2090faec786ecc7bd23b9e3b198 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/33/b703df2942bd2acbfaeb55cd5d4fca107ab2090faec786ecc7bd23b9e3b198 deleted file mode 100644 index 189f3fa..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/33/b703df2942bd2acbfaeb55cd5d4fca107ab2090faec786ecc7bd23b9e3b198 +++ /dev/null @@ -1,78 +0,0 @@ -I".

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new **effective yet optimization-friendly multi-contact model.** -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/34/ca0c00d5fb9e2a0b06324d770a7fa9b581bba44939f0a75fc69fc5fb73f874 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/34/ca0c00d5fb9e2a0b06324d770a7fa9b581bba44939f0a75fc69fc5fb73f874 deleted file mode 100644 index 2d0e8ec..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/34/ca0c00d5fb9e2a0b06324d770a7fa9b581bba44939f0a75fc69fc5fb73f874 +++ /dev/null @@ -1,93 +0,0 @@ -I"Ę

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation? YES! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

Thrilled to anounce our latest work: "Complementarity-Free Multi-Contact Modeling and Optimization", consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks (check out 🎥 below!), including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation, all with different objects. -

- Our method sets a new benchmark in dexterous manipulation: -
    -
  • 🎯 A 96.5% success rate across all tasks
    • -
  • ⚙️ Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • 🚀 Model predictive control running at 50-100 Hz for every task
    • -
-
- -
-
- Please check out the paper and code (fun guaranteed) -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check out project website and arXiv preprint breaf introduction of the paper: -
-
- -
-
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/35/22e9b76a2557f97877d88ad05ae93f5227bca8cd7001371ec6e93fcd3f2bf4 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/35/22e9b76a2557f97877d88ad05ae93f5227bca8cd7001371ec6e93fcd3f2bf4 deleted file mode 100644 index be631d5..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/35/22e9b76a2557f97877d88ad05ae93f5227bca8cd7001371ec6e93fcd3f2bf4 +++ /dev/null @@ -1,93 +0,0 @@ -I"č

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? YES! The key lies in our new "effective yet optimization-friendly multi-contact model". -

-

🔥 Thrilled to unveil our work: "Complementarity-Free Multi-Contact Modeling and Optimization", which consistently achieves state-of-the-art results across different challenging dexterous manipulation tasks, including fingertip 3D in-air manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation, all with different objects. Check out the demo below! -

- Our method sets a new benchmark in dexterous manipulation: -
    -
  • 🎯 A 96.5% success rate across all tasks
    • -
  • ⚙️ High manipulation accuracy: 11° reorientation error & 7.8 mm position error
    • -
  • 🚀 Model predictive control running at 50-100 Hz for all tasks
    • -
-
- -
-
- Check out our preprint, and try out our code (fun guaranteed). Here is a long demo: -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check out the project website, preprint, and a breaf introduction vide below. -
-
- -
-
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/36/7ddefc08590b07ea59f51ce23a340aa1b1dc2965bded34ad40e9ed6f8367c9 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/36/7ddefc08590b07ea59f51ce23a340aa1b1dc2965bded34ad40e9ed6f8367c9 deleted file mode 100644 index 633fa0c..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/36/7ddefc08590b07ea59f51ce23a340aa1b1dc2965bded34ad40e9ed6f8367c9 +++ /dev/null @@ -1,78 +0,0 @@ -I"A

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/36/fcbb4fdf5a9fefda9b0bbd50fbb6a60cc53ef0f83c93139c05f7a6d94aae04 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/36/fcbb4fdf5a9fefda9b0bbd50fbb6a60cc53ef0f83c93139c05f7a6d94aae04 deleted file mode 100644 index d2f9344..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/36/fcbb4fdf5a9fefda9b0bbd50fbb6a60cc53ef0f83c93139c05f7a6d94aae04 +++ /dev/null @@ -1,93 +0,0 @@ -I"ç

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? YES! The key lies in our new "effective yet optimization-friendly multi-contact model". -

-

🔥 Thrilled to unveil our work: Complementarity-Free Multi-Contact Modeling and Optimization, which consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks, including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation, all with different objects. Check out the demo below! -

- Our method sets a new benchmark in dexterous manipulation: -
    -
  • 🎯 A 96.5% success rate across all tasks
    • -
  • ⚙️ High manipulation accuracy: 11° reorientation error & 7.8 mm position error
    • -
  • 🚀 Model predictive control running at 50-100 Hz for all tasks
    • -
-
- -
-
- Check out our preprint, and try out our code(--- fun guaranteed!) Here is a long demo: -

-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check out the project website, preprint, and a breaf introduction vide below. -
-
- -
-
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/37/47c88d16dd35f01c021615df8cd3569959772f6c7276e41a1deaee39e6d684 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/37/47c88d16dd35f01c021615df8cd3569959772f6c7276e41a1deaee39e6d684 deleted file mode 100644 index 1661ab8..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/37/47c88d16dd35f01c021615df8cd3569959772f6c7276e41a1deaee39e6d684 +++ /dev/null @@ -1,87 +0,0 @@ -I" 

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
-
-
Aug 19, 2024
-
-

- Can model-based planning and control rival or even surpass reinforcement learning in challenging dexterous manipulation tasks? Absolutely! The key lies in our new effective yet optimization-friendly multi-contact model. -

-

Anounce our latest work: "Complementarity-Free Multi-Contact Modeling and Optimization." Our method consistently achieves state-of-the-art results across various challenging dexterous manipulation tasks (check out 🎥 below!), including 3D in-air fingertip manipulation, TriFinger in-hand manipulation, and Allegro hand on-palm manipulation—each with different objects. -

- - Our method sets a new benchmark in dexterous manipulation: -
    -
  • ✨ A 96.5% success rate across all tasks
    • -
  • 🎯 Precise manipulation: 11° reorientation error & 7.8 mm position error
    • -
  • ⚡ Model predictive control running at 50-100 Hz for every task
    • -
-
- -
-
-
- - - -


-
-
July 9 2024
-
-

- 🤖 Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -

- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -
- -
-
- Check a breaf introduction of the paper: -
-
- -
- Check out the project website and arXiv preprint. -
-
- - -
- -:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/37/7c76d2bbe62ced48c6254adfa7df0a0c5bb39fddfcde46c18c2cd42efd4b2a b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/37/7c76d2bbe62ced48c6254adfa7df0a0c5bb39fddfcde46c18c2cd42efd4b2a deleted file mode 100644 index 4f6c5b3..0000000 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/37/7c76d2bbe62ced48c6254adfa7df0a0c5bb39fddfcde46c18c2cd42efd4b2a +++ /dev/null @@ -1,95 +0,0 @@ -I"Ę

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

- - - -

- -
- -     - -     - -     - -     - -     - -
- -


- -

Recent Updates

- -

- -
- -
-
July 9 2024
-
-

- Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -
-
- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -

- -
-
-
- Check a breaf introduction of the paper: -
- -
- -
- -
- Check out the project website and arXiv preprint. -</div> -</div> - -



- - - -
-
July 9 2024
-
-

- Robots may be good at inferring a task reward from human feedback, but how about inferring safety boundaries from human feedback? In many cases such as robot feeding and liquid pouring, specifying user-comfortable safety constraints is more challenging than rewards. Our recent work, led by my PhD student Zhixian Xie, shows that this is possible, and can actually be very human-effort efficient! Our method is called Safe MPC Alignment (submitted to T-RO), enabling a robot to learn its control safety constraints with only a small handful of human online corrections! -
-
- Importantly, the Safe MPC Alignment is certifiable: providing an upper bound on the total number of human feedback in the case of successful learning of safety constraints, or declaring the misspecification of the hypothesis space, i.e., the true implicit safety constraint cannot be found within the specified hypothesis space. -

- -
-
-
- Check a breaf introduction of the paper: -
- -
- -
- -
- Check out the project website and arXiv preprint. -

-:ET \ No newline at end of file diff --git a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/7d/b576b791ba7c0bf2188dd43e556a5900ec5605a238c579b1206635436623b7 b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/3a/dc879be442ad7e0024fad342f7e488b1911e0f8f788542056e0e2ae8bc834f similarity index 74% rename from .jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/7d/b576b791ba7c0bf2188dd43e556a5900ec5605a238c579b1206635436623b7 rename to .jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/3a/dc879be442ad7e0024fad342f7e488b1911e0f8f788542056e0e2ae8bc834f index ad20334..d29b85c 100644 --- a/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/7d/b576b791ba7c0bf2188dd43e556a5900ec5605a238c579b1206635436623b7 +++ b/.jekyll-cache/Jekyll/Cache/Jekyll--Converters--Markdown/3a/dc879be442ad7e0024fad342f7e488b1911e0f8f788542056e0e2ae8bc834f @@ -1,4 +1,4 @@ -I"č

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include

+I"ü

This is Intelligent Robotics and Interactive Systems (IRIS) Lab! Our research focuses include