cdellin / or_benchmarks

Benchmarks to measure OpenRAVE's performance.

OpenRAVE benchmark tools

Benchmarks to measure OpenRAVE's performance.

Current Results

The following are the latest collision checking benchmark results:

Installation

The following packages must be installed:

• ros-hydro-moveit-core
• ros-hydro-moveit-ros-planning-interface
• libgoogle-perftools-dev

cmd> sudo apt-get install ros-hydro-moveit-core ros-hydro-moveit-ros-planning-interface \
libgoogle-perftools-dev

The following .rosinstall can be used to download the important packages into a catkin workspace:

- git: {local-name: or_benchmarks, uri: '[email protected]:personalrobotics/or_benchmarks'}
- git: {local-name: herbpy, uri: '[email protected]:personalrobotics/herbpy', version: 1.4.0}
- git: {local-name: prpy, uri: '[email protected]:personalrobotics/prpy', version: 0.4.0}
- git: {local-name: herb_description, uri: '[email protected]:personalrobotics/herb_description', version: 1.1.0}
- git: {local-name: or_urdf, uri: '[email protected]:personalrobotics/or_urdf', version: 0.2.0}
- git: {local-name: openrave_catkin, uri: '[email protected]:personalrobotics/openrave_catkin', version: 1.0.1}
- git: {local-name: ss_plotting, uri: '[email protected]:personalrobotics/ss_plotting', version: 0.2.0}
- git: {local-name: or_fcl, uri: '[email protected]:personalrobotics/or_fcl', version: 0.1.0}
- git: {local-name: pr-ordata, uri: '[email protected]:personalrobotics/pr-ordata', version: feature/pr_kitchen}

Kinematic Benchmarks

The kinematick benchmark script can be used to profile forward kinematic computation and Jacobian computation.

To profile forward kinematics:

cmd> rosrun or_benchmarks run_kinematics_benchmark.py --type fk

By default this will randomly sample 50,000 configurations for the right arm of HERB and compute forward kinematics for each of these configurations. To test a different manipulator, use the --manip flag:

cmd> rosrun or_benchmarks run_kinematics_benchmark.py --type fk --manip head

To profile Jacobian computation:

cmd> rosrun or_benchmarks run_kinematics_benchmark.py --type jacobian

Use the --help flag to see all options for the script:

cmd> rosrun or_benchmarks run_kinematics_benchmark.py --help

Collision Benchmarks

The collision benchmark script can be used to profile both self collision checking and environment collision checking.

To run all benchmarks and generated updated statistics simply run:

cmd> rosrun or_benchmarks run_all.py

This will run three sets of tests:

• Self collision - 20000 poses read from datasets/self_benchmark.test
• Collision against an empty environment - 20000 poses read from datasets/env_benchmark.test
• Collision against the pr_kitchen environment - 20000 poses read from datasets/env_benchmark.test Each test will be run using the ode, pqp and fcl collision checkers.

The script will generate a new set of .png files which compare timing information across the three collision checking engines. These new files will be written to the results directory. To save these results, and update this webpage, just commit the new files.

The run_all.py script utilizes the run_collision_benchmark.py script. Use run_collision_benchmark.py directly to run individual tests. The --help flag will show all options for the script:

cmd> rosrun or_benchmarks run_collision_benchmark.py --help