High-performance Racing on Unmapped Tracks using Local Maps

TL;DR

This work tackles high-performance autonomous racing in unmapped environments by introducing a local map framework that derives a concise representation of the visible track region from LiDAR data and feeds it to optimisation-based controllers. By combining local map extraction with two planning strategies—a two-stage optimisation and Model Predictive Contouring Control—the approach achieves fast lap times close to global-map methods while avoiding the need for a track map. In F1Tenth simulations, the local-map planner outperforms mapless baselines (8.8% faster than FTG and 3.22% faster than end-to-end networks) and is only marginally slower than global planners (≈3% on average), illustrating the viability of optimization-based control using local geometry for unmapped racing. The findings suggest that local maps can bridge the gap between map-based and mapless methods, enabling high-speed, map-free racing, with promising avenues for image-based cues, SLAM enhancement, and safe learning on unmapped tracks.

Abstract

Map-based methods for autonomous racing estimate the vehicle's location, which is used to follow a high-level plan. While map-based optimisation methods demonstrate high-performance results, they are limited by requiring a map of the environment. In contrast, mapless methods can operate in unmapped contexts since they directly process raw sensor data (often LiDAR) to calculate commands. However, a major limitation in mapless methods is poor performance due to a lack of optimisation. In response, we propose the local map framework that uses easily extractable, low-level features to build local maps of the visible region that form the input to optimisation-based controllers. Our local map generation extracts the visible racetrack boundaries and calculates a centreline and track widths used for planning. We evaluate our method for simulated F1Tenth autonomous racing using a two-stage trajectory optimisation and tracking strategy and a model predictive controller. Our method achieves lap times that are 8.8% faster than the Follow-The-Gap method and 3.22% faster than end-to-end neural networks due to the optimisation resulting in a faster speed profile. The local map planner is 3.28% slower than global methods that have access to an entire map of the track that can be used for planning. Critically, our approach enables high-speed autonomous racing on unmapped tracks, achieving performance similar to global methods without requiring a track map.

Figures (14)

• Figure 1: Local map racing pipeline; (1) receive LiDAR scan, (2) extract LocalMap, (3) calculate optimal trajectory.
• Figure 2: Map-based planners using a perception and optimisation-based strategy compared with mapless methods.
• Figure 3: Local map extraction: the track boundaries are identified and used to calculate a centre line and normal vectors of the visible region.
• Figure 4: Local map segment extraction (left) and a final local map (right) for segments of the AUT track. The purple arrow represents the car's pose.
• Figure 5: The two-stage planner generates an optimal trajectory that is tracked with a pure pursuit controller