Adaptive Constraint Planning for Racing Cars

Summary & Methodology Analysis

The research addresses the discrepancy between ideal trajectory planning and the reality of autonomous racecars at performance limits. Standard planners often rely on conservative spatial margins because they assume perfect model execution, which causes inefficient track usage. The authors introduce a closed-loop feedback mechanism where an observation module measures actual tracking deviations against an offline reference raceline. These measured errors are then used to dynamically adjust the spatial constraints of the plannable area for the Model Predictive Control (MPC) planner, which is an optimization-based control algorithm that predicts future system states to determine the best sequence of control inputs. By feeding these updated boundaries into the next lap planning cycle, the system iteratively optimizes the available track width. The paper implements this approach on the Yas Marina Circuit to demonstrate the improvement in performance compared to static models. The adaptive system utilizes four distinct heuristics to manage constraint updates: Linear, Curvature-scaled, Low-pass Filter, and Gaussian. While effective, the system faces specific limitations based on the chosen heuristic. The Low-pass Filter heuristic, which smooths out high-frequency noise in the error data, can lead to overcompensation and eventual controller instability. Conversely, the Gaussian heuristic provides more stable tracking behavior but tends to converge toward a performance limit that is less optimal than other methods in the set.