KnowRob

The purpose of KnowRob is to equip robots with explicit knowledge about the world. Originally, it was implemented using the Prolog programming language. In its second iteration, KnowRob is implemented in C++, but still supports Prolog for rule-based reasoning (see this page for more details).

The core of KnowRob is a shared library that implements a hybrid knowledge base. With hybrid, we mean that different reasoning engines can be combined in KowRob's query evaluation process. To this end, KnowRob defines a querying language and manages which parts of a query are evaluated by which reasoner or storage backend. Both reasoners and storage backends are configurable, and can be extended by plugins either written in C++ or Python. There are a few applications shipped with this repository including a terminal application that allows to interact with KnowRob using a command line interface.

For ROS integration, please refer to this repository.

Getting Started

These instructions will get you a copy of KnowRob up and running on your local machine.

Dependencies

The following list of software is required to build KnowRob:

• Redland and Raptor2
• spdlog
• fmt
• Eigen 3
• Boost >= 1.50 with the following components:
  • python
  • program options
  • serialization
• SWI Prolog >= 8.2.4
• mongo DB server >= 4.4 and libmongoc
• GTest

Optional Dependencies

Some features will only be conditionally compiled if the following dependencies are found:

• doxygen, for generating API documentation.

Installation

KnowRob uses CMake as build system. The following steps will guide you through the installation process. Assuming you have cloned the repository to ~/knowrob:

cd ~/knowrob
mkdir build
cd build
cmake -DCATKIN=OFF -DPYTHON_MODULE_LIBDIR="dist-packages" ..
make
sudo make install

The PYTHON_MODULE_LIBDIR option should be set to "site-packages" if you are using a non-Debian system. CATKIN=OFF is used to avoid the installation of unnecessary files.

You may further need to set the SWI_HOME_DIR environment variable to the installation location of swipl:

export SWI_HOME_DIR=/usr/lib/swi-prolog

Alternatively, you may clone the KnowRob repository into a ROS workspace and build it using catkin. Please refer to the ROS documentation for further information.

Plugin Installation

KnowRob attempts to load all plugins referred to in the active configuration file. To this end it will try to resolve relative paths using the following directories:

• ${SOURCE_PREFIX}/src
• ~/.knowrob
• ${INSTALL_PREFIX}/lib/knowrob (for shared libraries)
• ${INSTALL_PREFIX}/share/knowrob (for Python modules)

${SOURCE_PREFIX} is the directory where the source code is located, ${INSTALL_PREFIX} is the directory where the installation directory (usually "/usr/local"). Make sure to install plugins in one of these directories, or alternatively, refer to the plugin in the configuration file using an absolute path.

Development

Any IDE with proper CMake and C++ language support should be able to load the project. For example, you can use CLion or Visual Studio Code. But support for Prolog code is usually quite limited or not existent.

Configuration

KnowRob uses a configuration file to set up the knowledge base. Internally, boost's property tree is used to parse the configuration file. Hence, JSON is one of the supported formats. The configuration file specifies the storage backends, the reasoner, and the knowledge sources that are loaded into the knowledge base.

An example configuration is listed below:

{
  "data-sources": [
    {
      "path": "/path/to/owl/my-ontology.owl",
      "language": "owl",
      "format": "xml"
    }
  ],
  "data-backends": [
    {
      "name": "pl",
      "type": "Prolog:rdf_db"
    }
  ],
  "reasoner": [
    {
      "name": "pl",
      "type": "Prolog",
      "data-backend": "pl",
      "imports": [
        {
          "path": "/path/to/rules/my-rules.pl",
          "format": "prolog"
        }
      ]
    }
  ]
}

It configures KnowRob to use a builtin storage backend with type Prolog:rdf_db, and connects a Prolog reasoner to it (which is also a builtin reasoner type). The storage is populated with an OWL ontology in XML format, and the reasoner is extended with a set of Prolog rules via the "imports" configuration. The files are loaded from the specified paths, if these are provided as relative paths, KnowRob will attempt to resolve them relative to source, home, or installation directories.

For more information about storage backends, please refer to the Backends documentation, and for more information about reasoning, please refer to the Reasoner documentation. Additional examples of configuration files can be found in the settings and tests directories.

Launching

Being a shared library, KnowRob cannot be launched directly but only in the context of a program that uses it.

Using the Python Module

KnowRob provides a Python module that can be used to interact with the shared library. The module is generated during the build process and is installed in the Python module directory of the installation prefix (e.g., /usr/local/lib/python3/dist-packages). In the case of a ROS workspace, the module is installed in the devel directory.

To use the module, simply import it in your Python script:

import knowrob

The API of the Python module mirrors a part of the C++ API, and is designed to be as similar as possible. There is no separate API documentation for the Python module, as the API is (almost) the same as the C++ API (see API Documentation).

For more information on how to use the Python module, please refer to the Python Integration documentation, and the examples in the tests directory.

Using the Shared Library

Applications may choose to link with the shared library and use the provided C API (see API Documentation). The library is called libknowrob.so and is installed in the library directory of the installation prefix (usually /usr/local/lib). In the case of a ROS workspace, the library is installed in the devel/lib directory of the workspace. To link against the library, make sure the library installation directory is in the search path (LD_LIBRARY_PATH for Linux-based systems).

KnowRob further generates a pkg-config file that can be used to retrieve the necessary flags for compiling and linking against the library. The file is called knowrob.pc and is installed in the lib/pkgconfig directory of the installation prefix. To use it, add the following line to your CMakeLists.txt:

pkg_check_modules(KNOWROB REQUIRED knowrob)

Then you can use the KNOWROB_LIBRARIES and KNOWROB_INCLUDE_DIRS variables in your build.

In the case of a ROS workspace, simply do the following:

• add "knowrob" to the find_package and catkin_package calls in your CMakeLists.txt
• add "knowrob" to the depend fields in your package.xml

Using an interactive Terminal

KnowRob can also run as a standalone program knowrob-terminal that provides a command line interface. It can be launched as follows:

knowrob-terminal --config-file ~/knowrob/settings/prolog.json

The configuration file is a required argument, there is no fallback configuration file.

Once the terminal is up and running, you should see a greeting message and a prompt which looks like this:

Welcome to KnowRob.
For online help and background, visit http://knowrob.org/

?- 

Please refer to the Query documentation for the syntax of queries that can be typed into the terminal. Limited auto-completion is available. exit/0 will terminate the terminal.

Using a ROS Node

Alternatively, you can expose the KnowRob querying interface via a ROS node. The code for doing this is not part of this repository, but is available in the knowrob_ros repository.

Getting Familiar

Here we provide an overview about functionality of KnowRob.

Querying

The core of KnowRob is a querying interface that is built around a custom querying language. Its syntax is similar to Prolog, but it is more simplified and not Turing-complete like Prolog is.

For more information on querying in KnowRob, please have a look here.

Ontologies

KnowRob structures knowledge using ontologies. Ontologies are (formal) models of a domain that are used to describe the concepts in the domain and the relationships between them. In KnowRob, ontologies are usually represented as RDF knowledge graphs using the RDFS and OWL vocabularies.

Ontologies are organized in a hierarchy where each ontology is a specialization of another ontology. A common distinction is made between foundational (or top-level) ontologies, domain ontologies, and application ontologies. A foundational ontology fixes the basic concepts and relationships that are used across different domains. In KnowRob, we define a domain ontology for the robotics domain, and align it with a common foundational ontology. Applications can then import the domain ontology and extend it with application-specific concepts to cover the specific requirements of the application.

For more information on ontologies in KnowRob, please have a look here.

Triple Store and Data Access

Knowledge is represented in form of contextualized triples -- each subject-predicate-object triple has additional fields that contextualize the triple. A configurable storage backend is used to store and retrieve triples -- currently, triple stores based on Prolog, MongoDB and Redland are supported.

One important aspect in knowledge representation for robots is that a lot of knowledge is implicitly encoded in the control structures of the robot. Hence, one goal is to make this implicit knowledge explicit. This is done by mapping data to symbols in an ontology. In case of querying, this is often referred to as Ontology-based Data Access (OBDA).

For more information on storages in KnowRob, please have a look here.

Reasoning

KnowRob uses an ensemble of reasoners approach where inferences of different reasoners are combined into correlated knowledge pieces. The reason for choosing this approach is that there is no single method that is suited for every reasoning tasks. Instead, given a problem, one should decide what the most suitable method is to tackle it. Consequently, KnowRob can be configured to solve specific problems by loading corresponding reasoning modules that implement a common interface.

For more information on reasoning in KnowRob, please have a look here.

Further Information

More documentation can be found in the following pages:

• Terms
• Formulas
• Triples
• Ontologies
• Querying
• Backends
• Reasoner
• Python Integration

In addition, the following resources are available:

• API Documentation
• A blog and more wiki pages are available at knowrob.org