A Simulation Benchmark for Autonomous Racing with Large-Scale Human Data
Adrian Remonda, Nicklas Hansen, Ayoub Raji, Nicola Musiu, Marko Bertogna, Eduardo Veas, Xiaolong Wang
2024-07-24
Summary
This paper presents a new racing simulation platform based on the Assetto Corsa simulator, designed to test and improve autonomous driving algorithms using realistic scenarios and large amounts of human driving data.
What's the problem?
Research in autonomous racing has been limited by the high costs of real vehicles and the inaccuracies of open-source simulators. Many existing methods do not provide an effective way to evaluate how well self-driving cars can perform under challenging conditions, making it hard to develop reliable algorithms for racing.
What's the solution?
To address these challenges, the authors created a simulation platform that uses Assetto Corsa to provide a realistic environment for testing autonomous driving algorithms. This platform allows researchers to benchmark various algorithms, including reinforcement learning (RL) and Model Predictive Control (MPC), in a controlled setting. They also collected a comprehensive dataset from over one million laps driven by human drivers, which helps create more accurate and diverse testing scenarios for the algorithms.
Why it matters?
This research is significant because it provides a valuable tool for advancing the field of autonomous racing. By offering a realistic and accessible simulation environment, it enables researchers to develop and refine self-driving technologies more effectively. The large dataset of human driving behavior can help improve the performance of autonomous systems, making them safer and more reliable for real-world applications.
Abstract
Despite the availability of international prize-money competitions, scaled vehicles, and simulation environments, research on autonomous racing and the control of sports cars operating close to the limit of handling has been limited by the high costs of vehicle acquisition and management, as well as the limited physics accuracy of open-source simulators. In this paper, we propose a racing simulation platform based on the simulator Assetto Corsa to test, validate, and benchmark autonomous driving algorithms, including reinforcement learning (RL) and classical Model Predictive Control (MPC), in realistic and challenging scenarios. Our contributions include the development of this simulation platform, several state-of-the-art algorithms tailored to the racing environment, and a comprehensive dataset collected from human drivers. Additionally, we evaluate algorithms in the offline RL setting. All the necessary code (including environment and benchmarks), working examples, datasets, and videos are publicly released and can be found at: https://assetto-corsa-gym.github.io.