This repo contains the code for the paper, "From Navigation to Racing: Reward Signal Design for Autonomous Racing"
- We evaluate different reward signals on F1/10th autonomous cars
- We test the reward signals using a planning, perception and control navigation stack with a minimum curvature global planner and the Reference Modification local planner
- The repo contains two scripts to generate the results which are located in the
TestingScripts
folder - The
TrainVehicles.py
Script will train the vehicles with the different reward signals - The
TestVehilces.py
script, evaluated the vehicles. - All the training and evaluation is done on the porto F1/10th race track.
- No racing reward
- Centerline progress reward
- Global plan progress reward
- Centerline cross-track, heading error reward
- Global plan cross-track, heading error reward
- Steering punishment reward
- Requirements:
- PyTorch
- Numpy
- Matplotlib
- casadi
- numba
- scipy
- Installation
- clone the repo onto your computer
- navigate into the folder,
cd RewardSignalDesign
- install it using pip
python3 -m pip install -e .
- Built on Linux Ubuntu system (20.04.2 LTS) using Python v3.8.5
If you have found our work helpful, please cite as:
@inproceedings{evans2021reward,
title={Reward signal design for autonomous racing},
author={Evans, Benjamin and Engelbrecht, Herman A and Jordaan, Hendrik W},
booktitle={2021 20th International Conference on Advanced Robotics (ICAR)},
pages={455--460},
year={2021},
organization={IEEE}
}