[Paper Review] TUM autonomous motorsport: An autonomous racing software for the Indy Autonomous Challenge

Johannes Betz, Tobias Betz+17 more|arXiv (Cornell University)|May 31, 2022

Autonomous Vehicle Technology and Safety4 citations

TL;DR

This paper presents the TUM Autonomous Motorsport software stack for the Indy Autonomous Challenge, featuring a modular, simulation-driven development pipeline enabling high-speed autonomous racing on oval tracks. The team achieved first place in Indianapolis and second in Las Vegas, reaching speeds up to 270 km/h and accelerations of 28 m/s², demonstrating robust performance at vehicle limits through integrated perception, localization, planning, and control algorithms with extensive real-world and simulation testing.

ABSTRACT

For decades, motorsport has been an incubator for innovations in the automotive sector and brought forth systems like disk brakes or rearview mirrors. Autonomous racing series such as Roborace, F1Tenth, or the Indy Autonomous Challenge (IAC) are envisioned as playing a similar role within the autonomous vehicle sector, serving as a proving ground for new technology at the limits of the autonomous systems capabilities. This paper outlines the software stack and approach of the TUM Autonomous Motorsport team for their participation in the Indy Autonomous Challenge, which holds two competitions: A single-vehicle competition on the Indianapolis Motor Speedway and a passing competition at the Las Vegas Motor Speedway. Nine university teams used an identical vehicle platform: A modified Indy Lights chassis equipped with sensors, a computing platform, and actuators. All the teams developed different algorithms for object detection, localization, planning, prediction, and control of the race cars. The team from TUM placed first in Indianapolis and secured second place in Las Vegas. During the final of the passing competition, the TUM team reached speeds and accelerations close to the limit of the vehicle, peaking at around 270 km/h and 28 ms2. This paper will present details of the vehicle hardware platform, the developed algorithms, and the workflow to test and enhance the software applied during the two-year project. We derive deep insights into the autonomous vehicle's behavior at high speed and high acceleration by providing a detailed competition analysis. Based on this, we deduce a list of lessons learned and provide insights on promising areas of future work based on the real-world evaluation of the displayed concepts.

Motivation & Objective

• To develop a high-performance, modular autonomous racing software stack capable of operating at the physical limits of a race car.
• To enable safe and reliable autonomous racing on high-speed oval tracks through rigorous simulation and real-world testing.
• To establish a scalable and extensible software development workflow that supports rapid iteration and validation of autonomous driving algorithms.
• To derive actionable insights from real-world competition data to guide future research in autonomous vehicle safety and scalability.
• To promote open research by sharing the software stack and development pipeline as a public resource for the autonomous vehicle community.

Proposed method

• Design of a modular software architecture integrating perception, localization, prediction, planning, and control modules using ROS 2 for real-time communication.
• Implementation of a high-fidelity simulation environment based on CARLA for pre-deployment testing and algorithm validation.
• Development of a multi-stage testing pipeline combining simulation, hardware-in-the-loop (HIL), and real-world track testing at Lucas Oil Raceway, Indianapolis Motor Speedway, and Las Vegas Motor Speedway.
• Employment of advanced perception algorithms using YOLOv5 for object detection and a learned motion model for dynamic obstacle prediction.
• Application of nonlinear model predictive control (NMPC) for trajectory planning with real-time optimization under kinematic and dynamic constraints.
• Use of a hybrid localization approach combining GPS/IMU dead reckoning with map-based refinement using a pre-mapped track centerline and lane markings.

Experimental results

Research questions

• RQ1 How can a modular, simulation-validated software stack be developed to enable autonomous racing at speeds approaching the vehicle's physical limits?
• RQ2 What role does a comprehensive simulation and testing pipeline play in ensuring robustness and safety during high-speed autonomous racing?
• RQ3 How do perception, prediction, and control algorithms interact under extreme dynamic conditions such as 28 m/s² accelerations and 270 km/h speeds?
• RQ4 What are the key challenges in scaling autonomous racing to multi-vehicle scenarios, and how can they be addressed through software design and simulation?
• RQ5 What lessons from real-world autonomous racing are transferable to public road deployment and future urban autonomy applications?

Key findings

• The TUM software stack enabled the vehicle to reach a peak speed of 270 km/h and a maximum acceleration of 28 m/s² during the Las Vegas passing competition, demonstrating operation near the vehicle's physical limits.
• The team secured first place in the single-vehicle Indianapolis Motor Speedway competition and second place in the multi-vehicle passing competition at Las Vegas Motor Speedway.
• The simulation-driven development pipeline significantly reduced real-world testing risks and enabled iterative refinement of perception, prediction, and control algorithms.
• The integration of NMPC-based trajectory planning with real-time obstacle prediction allowed for safe and aggressive overtaking maneuvers under high-speed conditions.
• The modular, ROS 2-based architecture enabled efficient software maintenance and rapid deployment across different competition formats and vehicle configurations.
• Competition analysis revealed that robust handling of uncertainty—especially multimodal prediction and sensor noise—was critical for high-performance autonomous racing.