On-line POMDP Planning Toolkit (OPPT)

OPPT

On-line POMDP Planning Toolkit (OPPT) is a C++ software-toolkit for approximating POMDP solutions on-line. To ease implementation of POMDP models, OPPT allows users to specify the state, action, and observation spaces via a configuration file, and uses a plug-in architecture to implement the core components of a POMDP, that is, the transition, observation and reward functions.

Installing the dependencies

One-line installation script

OPPT provides an installation script that installs all the dependencies:

cd <oppt_root_folder>
chmod +x install_dependencies.sh && ./install_dependencies.sh --use-ros

This script will install the dependencies inside the /usr/local folder, hence, it will ask you for your sudo password. It has been tested on a clean Ubuntu 16.04, 18.04 and 20.04 installation. If you wish to not install the optional ROS dependency, run the above command without the "--use-ros" option.

Step-by-step installation

1. Install the core dependencies

    sudo apt-get install build-essential git cmake mercurial pkg-config libboost-all-dev libtinyxml-dev libeigen3-dev libassimp-dev libfcl-dev
   

2. Install Gazebo and its dependencies by following the instructions according to the Gazebo website. Both gazebo<GAZEBO_VERSION> and libgazebo<GAZEBO_VERSION>_dev are required, where <GAZEBO_VERSION> depends on your Ubuntu version.

3. Download and install libspatialindex-1.8.5

    wget http://download.osgeo.org/libspatialindex/spatialindex-src-1.8.5.tar.gz
    tar zxfv spatialindex-src-1.8.5.tar.gz
    cd spatialindex-src-1.8.5
    mkdir build && cd build
    cmake -DCMAKE_INSTALL_PREFIX=/usr/local ..
    make -j2 && sudo make install
   

4. (Optional) If you want to use the GUI provided in OPPT or the Inverse Kinematics functionality, you need to have a working installation of ROS (at least the ROS-base stack along with the rviz, kdl-parser and trac-ik stack). Furthermore your ROS environment has to be set up (please refer to http://wiki.ros.org//Installation/Ubuntu section 1.6). If you do not have ROS installation, you can install it using

    sudo sh -c 'echo "deb http://packages.ros.org/ros/ubuntu $(lsb_release -sc) main" > /etc/apt/sources.list.d/ros-latest.list'
    sudo apt-get update
    sudo apt-get install ros-<ROS DISTRIBUTION>-ros-base ros-<ROS DISTRIBUTION>-rviz ros-<ROS DISTRIBUTION>-kdl-parser ros-<ROS DISTRIBUTION>-trac-ik ros-<ROS DISTRIBUTION>-trac-ik-lib
   

where <ROS_DISTRIBUTION> is the ROS distribution corresponding to your Ubuntu version.

Setting-up ROS Kinect (optional)

If you have a working ROS installation, make sure that the ROS environment is set-up properly. For a clean ROS installation, this is done via

source /opt/ros/<ROS DISTRIBUTION>/setup.sh

If you fail to do this, you won't be able to use the viewer.

Building and installing OPPT

cd <oppt_root_folder>/src/
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=<BUILD_TYPE> -DCMAKE_INSTALL_PREFIX=<oppt_install_location> ..
make && make install

OPPT supports three build types: Debug, RelWithDebInfo and Release. The default build type is RelWithDebInfo

Configuring the OPPT runtime environment

In order to obtain resources such as SDF models, plugins, etc., OPPT uses a filesystem API

oppt::FileExists("filename")
oppt::FindFile("filename")

which locates resources inside folders specified in the

OPPT_RESOURCE_PATH

environment variable. To configure your current environment, either call

source <oppt_install_location>/share/oppt/setup.sh

or add this line to your bash environment file

This will add the

<oppt_install_location>/share/oppt/models
<oppt_install_location>/share/oppt/plugins

folders to the resource environment variable. If you want to add additional folders, use

export OPPT_RESOURCE_PATH=$OPPT_RESOURCE_PATH:<additional_folder>

Quick start

After the OPPT environment has been configured, navigate to the

<oppt_root_folder>/bin

directory. Here you should see the 'abt' executable. The problem configuration files for the problems that ship with OPPT are inside the

<oppt_root_folder>/cfg

directory. To run the ABT solver for any of the problems, run

./abt --cfg <oppt_root_folder>/cfg/<ProblemConfigurationFile>.cfg

Note that the path to the problem configuration file must be an absolute path.

If you want a visualization of the simulation runs, open a separate terminal and run

opptViewer

BEFORE running the ABT solver.

Documentation

The OPPT HTML documentation can be found inside the

<oppt_root_folder>/docs/html

folder

Changelog

v0.1

• Initial public release

v0.2

• Inside the modified ODE physics engine used by Gazebo, a cumulated joint angle parameter per joint is maintained during a physics update. This joint angle parameter is not being reset when setting the Gazebo world state, causing non-deterministic results. This is fixed by resetting the parameter before each physics engine update.
• Added virtual setUserData and getUserData the the OPPTPlugin. These methods can be overwritten and used to share objects amongst different plugins.
• Build the core OPPT framework as a shared library that solvers link against. This reduces compilation time when many solvers are being compiled.
• Added APIs to the Robot class that allows the user to set custom collision functions
• Added support for trac_ik inverse kinematics engine

v0.3

• Added support for Gazebo9.
• Many bugfixes and performance improvements.

v0.4

• Remove the necessity to define the environment and the robot in separate SDF files. The user now has to only specify the robot name in the configuration file.

v0.5

• When running multiple OPPT processes on the same machine, make sure that only the first one can use the viewer.
• Fixed a bug where state vectors are not correctly applied when the state description contains linkPositionX, linkPositionY and linkPositionZ variables.
• Make the specification of the 'planningEnvironmentPath' and 'executionEnvironmentPath' in the problem configuration file optional. This allows for the definition of general POMDP problems that are not dependend on geometrical or physical robot/environment models. In case these two options are not specified, the physics engine and the collision checking won't be initialized.
• Fixes in installation script.

Publication

Marcus Hoerger, Hanna Kurniawati, and Alberto Elfes. A Software Framework for Planning under Partial Observability. Proc. IEEE/RSJ Int. Conference on Intelligent Robots (IROS). 2018. [PDF].

Bibtex

@inproceedings{hoerger18:OPPT,
  title={A Software Framework for Planning Under Partial Observability},
  author={Hoerger, Marcus and Kurniawati, Hanna and Elfes, Alberto},
  booktitle={Proc. IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  pages={1--9},
  year={2018},
  organization={IEEE}
}

Contact

For questions, bug reports or suggestions please contact Marcus Hoerger, E-Mail: [email protected]