er.autopilot 1.0: The Full Autonomous Stack for Oval Racing at High Speeds

TL;DR

This work presents er.autopilot 1.0, the full autonomous racing stack used by TII EuroRacing in the Indy Autonomous Challenge to achieve high-speed oval racing with obstacle avoidance and overtakes. It integrates a multi-sensor perception system, a Kalman-filter-based sensor fusion framework, Frenet-based planning with a global minimum-time planner, and a high-speed model-predictive controller derived from a Dymola VeSyMA multi-body model, validated through simulation and two competitive events. The authors provide extensive results on localization accuracy, perception performance, and planning/control behavior, and extract lessons on GNSS reliability, model uncertainty, and safety thresholds, reporting second- and third-place finishes at IMS and CES respectively. The work offers practical insights for edge-enabled autonomous racing and informs future enhancements toward GNSS-denied operation, more aggressive yet safe local planning, and tighter integration between planning and control for multi-agent racing scenarios.

Abstract

The Indy Autonomous Challenge (IAC) brought together for the first time in history nine autonomous racing teams competing at unprecedented speed and in head-to-head scenario, using independently developed software on open-wheel racecars. This paper presents the complete software architecture used by team TII EuroRacing (TII-ER), covering all the modules needed to avoid static obstacles, perform active overtakes and reach speeds above 75 m/s (270 km/h). In addition to the most common modules related to perception, planning, and control, we discuss the approaches used for vehicle dynamics modelling, simulation, telemetry, and safety. Overall results and the performance of each module are described, as well as the lessons learned during the first two events of the competition on oval tracks, where the team placed respectively second and third.

Figures (26)

• Figure 1: Dallara AV-21
• Figure 2: Diagram block with the software modules of er.autopilot. The modules marked with an asterisk are connected with all the other modules.
• Figure 3: Top view of the LiDAR map obtained for the LVMS circuit. The color used for the points in the cloud is determined by the intensity value of each point.
• Figure 4: Perception scheme of er.autopilot.
• Figure 5: Results of the clustering BEV pipeline. The algorithm employs a low-resolution representation of the BEV, depicted on the left, where the white lines indicate the track walls, and their respective clusters are ignored. Instead, the rectangles represent the clusters, color-coded based on their local tracker, and the red arrows indicate their speed vectors. On the right side, the clusters are projected back onto the cloud.