GripMap: An Efficient, Spatially Resolved Constraint Framework for Offline and Online Trajectory Planning in Autonomous Racing

TL;DR

GripMap tackles the problem of location-dependent tire-road grip in autonomous racing by introducing a Frenet-frame, spatially resolved constraint framework. It stores a local grip-scaling factor θ_{ij} in a dense M×N grid and enables efficient O(1) lookups via perfect hashing, allowing both offline raceline optimization and high-frequency online planning to account for locally varying grip. The approach yields a 5.2% lap-time improvement on Yas Marina, maintains dynamic feasibility, and reduces risk in multi-vehicle interactions while incurring only about 0.77% additional runtime. By bridging offline and online planning with interpretable, location-specific data, GripMap lays the groundwork for real-time adaptation and broader applications beyond autonomous racing.

Abstract

Conventional trajectory planning approaches for autonomous vehicles often assume a fixed vehicle model that remains constant regardless of the vehicle's location. This overlooks the critical fact that the tires and the surface are the two force-transmitting partners in vehicle dynamics; while the tires stay with the vehicle, surface conditions vary with location. Recognizing these challenges, this paper presents a novel framework for spatially resolving dynamic constraints in both offline and online planning algorithms applied to autonomous racing. We introduce the GripMap concept, which provides a spatial resolution of vehicle dynamic constraints in the Frenet frame, allowing adaptation to locally varying grip conditions. This enables compensation for location-specific effects, more efficient vehicle behavior, and increased safety, unattainable with spatially invariant vehicle models. The focus is on low storage demand and quick access through perfect hashing. This framework proved advantageous in real-world applications in the presented form. Experiments inspired by autonomous racing demonstrate its effectiveness. In future work, this framework can serve as a foundational layer for developing future interpretable learning algorithms that adjust to varying grip conditions in real-time.

Figures (8)

• Figure 1: GripMap structure in the Frenet frame, which is centered around a reference line (centerline). The GripMap cells are discretized in the Frenet frame with constant step widths $\Delta s$ and $\Delta n$.
• Figure 2: Overview of our GripMap Framework: We discretize the track geometry in the Frenet frame and combine it with a baseline vehicle model. Each GripMap cell locally refines this model, in our case by applying a scaling factor $\theta$ to a point-mass model governed by g-g-g-v constraints. Together, these elements form the GripMap, which serves as the vehicle-dynamics constraint in our planning modules.
• Figure 3: GripMap discretization and corresponding scaling values for Turn 5 of the Yas Marina Circuit in Abu Dhabi. The highest grip is observed on the racing line, with decreasing grip values further away, reflecting the increased accumulation of debris and dirt on less frequently driven areas of the track.
• Figure 4: Velocity profiles for an Abu Dhabi lap obtained via offline raceline optimization using a uniform 75% global scaling factor versus the final spatially resolved GripMap (ranging between 75% and 100%). The locally varying scaling factors lead to noticeable differences in braking points and corner speeds, highlighting the impact of utilizing spatially variant dynamic constraints.
• Figure 5: Beginning of a spin-out during the IAC 2023 final at Las Vegas Motor Speedway. Dust kicked up by the tires on the outside line of a banked turn is clearly visible. This dust on the driving surface reduced grip availability, and without a spatial resolution of the vehicle dynamic constraints, the planning algorithm generated a dynamically unfeasible trajectory. The control algorithm, attempting to follow this unfeasible trajectory, ultimately led to the observed spin.