arxivcs.ROcs.AIcs.LG2026-07-08
Shift & Drift: A Zero-Shot Benchmark for Generalizable and Robust Autonomous Driving Motion Planning
Alessandro Canevaro, Hang Yu, Julian Schmidt, Peizheng Li, Silvan Lindner, Wilhelm Stork, et al.
While closed-loop motion planners trained on large-scale, object-level datasets, e.g., nuPlan, demonstrate strong in-distribution (ID) performance, their generalization to novel urban topologies and recovery mechanisms following execution perturbations remain under-explored. To a…