From Zero to High-Speed Racing: An Autonomous Racing Stack

TL;DR

The paper tackles the challenges of high‑speed autonomous racing by presenting the Autonomous Race Stack (ARS), a modular software-hardware stack evolved through three iterations (ARS1–ARS3) and validated on oval and road-course tracks at speeds up to 260 km/h. It details architectural progress, subsystem evaluations, and a high‑speed multi-sensor dataset, highlighting advances in GNSS/IMU fusion, radar LiDAR perception, and stochastic control (MPPI) for solo and multi‑vehicle racing. Key contributions include a systematic evolution of the ARS, comprehensive performance analyses across different track types, and practical insights from a real‑world spin incident and perception challenges. The work demonstrates real‑world viability of autonomous racing stacks, informs design tradeoffs for perception, localization, and control under high dynamics, and provides data to benchmark high‑speed autonomous racing systems.

Abstract

High-speed, head-to-head autonomous racing presents substantial technical and logistical challenges, including precise localization, rapid perception, dynamic planning, and real-time control-compounded by limited track access and costly hardware. This paper introduces the Autonomous Race Stack (ARS), developed by the IU Luddy Autonomous Racing team for the Indy Autonomous Challenge (IAC). We present three iterations of our ARS, each validated on different tracks and achieving speeds up to 260 km/h. Our contributions include: (i) the modular architecture and evolution of the ARS across ARS1, ARS2, and ARS3; (ii) a detailed performance evaluation that contrasts control, perception, and estimation across oval and road-course environments; and (iii) the release of a high-speed, multi-sensor dataset collected from oval and road-course tracks. Our findings highlight the unique challenges and insights from real-world high-speed full-scale autonomous racing.

Figures (9)

• Figure 1: IU-Luddy's IAC AV24 racing vehicle at LVMS.
• Figure 2: System architecture of ARS2 with core modules' details.
• Figure 3: Front Radar Pointcloud.
• Figure 4: (Top) Pointcloud mapping at LVMS with poses estimated by dynamics-satnav-inertial fusion. (Bottom) GPS dropout simulations on track with zoomed-in view.
• Figure 5: Comparison of estimated and ground-truth opponent states: (a) and (c) depict the UKF-estimated XY position ($x_{est}$, $y_{est}$) alongside the ground truth ($x_{gt}$, $y_{gt}$) from shared opponent localization, where the position error is defined as the Euclidean distance between them and scaled by a factor of 100 for clarity; (b) and (d) present the estimated velocity ($v_{est}$) versus the ground truth ($v_{gt}$), along with the absolute velocity error.