HyperPlan: Motion Planning Hyperparameter Optimization

HyperPlan is a tool for automatic selection of a motion planning algorithm and its parameters that optimize some performance metric over a given set of problems.

It uses HpBandSter as the underlying optimization package.

Installation

Run the following commands:

sudo apt-get install python3.8-dev python3.8-venv
python3.8 -m venv venv
source venv/bin/activate
pip install --isolated -r requirements.txt

OMPL.app benchmarking

Build a recent version of OMPL.app somewhere. Make sure that ompl_benchmark is somewhere in the $PATH.

MoveIt benchmarking

TODO

Robowflex benchmarking

1. Install ROS Melodic and these build tools:

     sudo apt-get install python-wstool python-catkin-tools
   

2. Check out the code:

     export CATKIN_WS=~/ws_hyperplan
     mkdir -p $CATKIN_WS/src
     cd $CATKIN_WS/src
     wstool init .
     git clone [email protected]:KavrakiLab/hyperplan.git
     wstool merge $CATKIN_WS/src/hyperplan/hyperplan.rosinstall
     wstool update
   

3. Configure and build the code:

     cd $CATKIN_WS
     catkin config --extend /opt/ros/$ROS_DISTRO \
       --cmake-args -DCMAKE_BUILD_TYPE=Release -DCMAKE_EXPORT_COMPILE_COMMANDS=ON
     rosdep install -y --from-paths src --ignore-src --rosdistro $ROS_DISTRO
     catkin build
   

4. Source the workspace.

     source $CATKIN_WS/devel/setup.bash
   

Running the optimization

Below are some examples on how to run the command line tool to optimize for different objectives using different backends.

OMPL.app benchmarking

Find a planner configuration that optimizes speed of geometric planning:

./examples/speed.sh

Find a planner configuration that optimizes speed of kinodynamic planning:

./examples/speed_kinodynamic.sh

Find a planner configuration that optimizes convergence rate of asymptotically (near-)optimal geometric planning:

./examples/convergence.sh

Type ./scripts/hyperplan.py --help to see all options. See the shared directory for the result files. See the scripts ./scripts/analysis.py and ./scripts/hyperplanvis.{py,R} for examples of how to perform some basic analysis of the results.

MoveIt

TODO

Docker

Using a docker image makes it easier to run multiple HyperPlan workers on the same machine. You can download a docker image of the version that was used for the IROS 2021 paper via these commands:

docker pull mmoll/hyperplan:latest
docker tag mmoll/hyperplan:latest hyperplan:latest

You can build your own hyperplan of the current version of the code like so:

catkin build docker-hyperplan

Check out ./examples/robowflex-docker.sh for an example of how to use the Docker image.