Fast and Modular Autonomy Software for Autonomous Racing Vehicles

TL;DR

This work presents a modular, fast ARV software stack designed for high-speed racing in the Indy Autonomous Challenge, demonstrated on the Dallara AV-21 platform. It integrates a multi-modal perception pipeline (Camera + LiDAR) with a tracking system, a planning module leveraging offline raceline banks and online action primitives, and a fast, LQR-based lateral control with a simple longitudinal controller. The paper reports both successes (high-speed racing up to ~150 mph, reliable opponent tracking at long range, and successful multi-agent passes) and failures (perception/tracking outages, spinouts due to tire temperature, and planner bugs), extracting actionable lessons to improve sensing reliability, tire modeling, and planning robustness. Overall, the study highlights modularity and speed as core design principles, enabling rapid experimentation and continual improvement across seasons, with practical implications for real-world ARV autonomy in dynamic, competitive environments.

Abstract

Autonomous motorsports aim to replicate the human racecar driver with software and sensors. As in traditional motorsports, Autonomous Racing Vehicles (ARVs) are pushed to their handling limits in multi-agent scenarios at extremely high ($\geq 150mph$) speeds. This Operational Design Domain (ODD) presents unique challenges across the autonomy stack. The Indy Autonomous Challenge (IAC) is an international competition aiming to advance autonomous vehicle development through ARV competitions. While far from challenging what a human racecar driver can do, the IAC is pushing the state of the art by facilitating full-sized ARV competitions. This paper details the MIT-Pitt-RW Team's approach to autonomous racing in the IAC. In this work, we present our modular and fast approach to agent detection, motion planning and controls to create an autonomy stack. We also provide analysis of the performance of the software stack in single and multi-agent scenarios for rapid deployment in a fast-paced competition environment. We also cover what did and did not work when deployed on a physical system the Dallara AV-21 platform and potential improvements to address these shortcomings. Finally, we convey lessons learned and discuss limitations and future directions for improvement.

Figures (38)

• Figure 1: Autonomy for AVs and ARVs is typically distilled into three broad tasks: Sense, Think, and Act. Sense is measuring the state of the environment. Think is deciding the best course of action to take. Act is executing on that course of action.
• Figure 2: Example of perception challenges faced by Autonomous Racing Vehicles (ARVs). (Left) Speed deltas between agents can be enough to cause motion blur. (Right) As the other agent passes, their acceleration is kicking up significant dirt, dust, and debris, which challenges LiDAR detection. Overall, sensor noise impairs robust detection and localization of opponent ARVs.
• Figure 3: DARPA Grand and Urban Challenges, two prize competitions for American autonomous vehicles. The Challenges were created to spur the development of autonomous vehicle technologies capable of completing a substantial off-road course, and later an urban environment course, within a limited time. Images from grand_challengeurban_challenge.
• Figure 4: Racing lanes for the Autonomous Challenge @CES at the Las Vegas Motor Speedway (January 2022/January 2023)
• Figure 5: Sensors on the AV-21. Six cameras and three LiDARs provide redundant $360^{\circ}$ coverage and over $200m$ of sensing range.