Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Reba Score #32

Open
wants to merge 32 commits into
base: main
Choose a base branch
from
Open

Conversation

sammutiti
Copy link

No description provided.

longnguyen2706 and others added 30 commits June 1, 2023 00:14
* fix installation issue. writing human lying, visualize pose and add scanned smpl files

* installed on hrl ghost. made some progress on mapping - leg seems to be correct

* added visualize script

* can map smpl joint angle to pybullet joint angle for urdf model

* able to load, modify and save new urdf. still trying to figure out how to scale human part

* more realistic human urdf. play with human mesh

* refactor human

* able to generate mesh for body part from smpl. able to make human from mesh. need to fix bug

* get a urdf working with human mesh

* able to offset joint in urdf, but current joint is set to child position - bug

* legs and hips joint working

* it works

* clean code. fix bug with urdf. fix euler conversion for pose angle

* add joint limit to urdf editor. create human_comfort_env. implement forward ik and manipubility func. implement train with CMAES (can run but seems to have bug)

* remove all redundant radius and mass for fake links. fix self collision issue

* further fix self-collision and human fly away issue. code clean up

* self collision working

* WIP - using kinpy for FK

* update axis rot for joint in urdf

* using pybullet ik with limited joint chain seems to work

* CMAES works with direct angle set

* clean up

* further clean up

* delete unnecessary files

* further clean

* fix issue
* add contact detection & closest point detection. contact need simStep() or using newer version of pybullet to call performCollisionDetection()

* update to latest pybullet. now could call API to detect contact. inverse dynamic also works

* clean up code

* calc potential energy change. add env collision detection

* fix global orient. debug and fix model flyaway (trick is just dont let body penetrade the floor and set contact damping high enough

* fix self collision. refactor the code

* further fix self collision

* add fibonacci point sampling

* refactor to train and render for various smpl angle. wrote bash script for parallel training

* add even point sampling method. refactor

* Delete test_mesh_bk.urdf
* remove target points that have collision. also train with energy

* add logger. clean code

* fix torque calculation. train with torque

* fix self collision hand - spine 4. fix render result

* change collision damping coefficient
* train with collision on. add early stopping once collision or no further change happen on simstep

* init CMAES with valid range. add reset human method

* refactor - add simulate-collision modee

* use CMAES to find max dynamic value. scale all dynamic wrt to max val
* can do ik with PR2, but not the stretch

* ik works with stretch and pose 8

* able to place robot on left/ right and do ik with collision avoidance. ik success is hit and miss

* toc ik works

* some refactoring. train ik toc works with various pose

* separate robot ik to a new mode
…ension optimization for cup, removed cane cost parameters
#5)

* added reba score calculation to human_comfort and temporarily disabled stretch

* added get_reba_score to human_urdf and added that to the train.py cost_fn

* added optional perp wrist and testing for cane, pill, and bottle

* added left hand capabilities and hand choice in reba score

* implemented object specification to modify cost optimization for pill

* added cup capbilities, upper hand selection for cup handover, 2nd dimension optimization for cup, removed cane cost parameters

* quick fix: support left hand render. also, fix collision with env

* added ray cast functions and troubleshooting ray cast error

* added cane capability and raytests

* added cleaned cost function and renamed human_urdf functions

* changed cup and pill optimzation to be vector based

* reset meshes and urdf to main

* clean up

* further clean up

---------

Co-authored-by: sammutiti <samanthahmutiti@gmail.com>
)

added stretch dex with r,p,y gripper

object orientation in hand is correct. need to clean the code

minor tweak and clean up

add collision object around hand with size and offset from hand bounding box. almost work

commit before switching branch

offset seems to be wrong

add collision object works. clean the code

clean up
* refactor: render env based on task and handover object

* minor change. re-enable search for max val
* refactor: render env based on task and handover object

* minor change. re-enable search for max val

* clean up train

* change handover target to 5cm away from the palm. also, change collision cost to COST * num of collision

* fix robot stuck when do not have valid ik solution
longnguyen2706 and others added 2 commits July 31, 2023 15:07
* refactor: render env based on task and handover object

* minor change. re-enable search for max val

* clean up train

* change handover target to 5cm away from the palm. also, change collision cost to COST * num of collision

* fix robot stuck when do not have valid ik solution

* change struture of saved actions.pkl file to support multiple handover object. update train.sh and readme
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants