PyRacecarSimulator

Simulator for simulating a F1-tenth car. Itegrated with ROS and uses range_libc for lidar simulations. Implemented to facilitate a Monte-Carlo Tree Search agent.

Depends

ROS

• ROS melodic
• Ackermann Messeges
• LaserScan Messeges
• Pose Messges
• Geometry Messages
• RViz

Other libs

• range_libc
• Tensorflow 1.13, with cudaNN 7.3
• Numpy

Installing

Put the project folder into a ros workspace and run,

catkin_make
source devel/setup.bash

Usage

Tune the simulation and the agents using the param.yaml file. To change policy for MCTS, edit mcts.py and swap out generateActionXX().

Simulation only

Build racecar Build range_libc

To run the simulation as a ROS node run,

roslaunch PyRacecarSimulator simulate.launch

Simulation with MCTS

Build the libraries,

bash racecar/compile.sh
bash range_libc/pywrapper/compile_with_cuda.sh
bash followthegap/compile.sh

To run the simulation as a ROS node run,

roslaunch PyRacecarSimulator simulate.launch

Then run the MCTS agent,

roslaunch PyRacecarSimulator mcts_driver.launch

TODO

• Add multi-agent capabilities
• Add support for incremental map updates
• Try other roll-out policies

Acknowledgements

Thanks to the code below we were able to create this project.

• mit-racecar
• range_libc
• racecar_simulator