Shift & Drift: A Zero-Shot Benchmark for Generalizable and Robust Autonomous Driving Motion Planning
arXiv:2607.07844v1 Announce Type: cross Abstract: 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 address this, we present Shift & Drift, a novel dual-track benchmark designed to rigorously stress-test motion planners across two critical axes of distribution shift: (1) The Semantic Shift Track leverages a novel conversion pipeline that transforms the aerial, DeepScen
The proliferation of advanced AI in safety-critical applications like autonomous driving necessitates more robust and generalizable evaluation benchmarks to ensure real-world effectiveness.
Improving the generalizability and robustness of autonomous driving motion planners is crucial for widespread adoption and the safety of AI-driven vehicles, accelerating the path to commercially viable self-driving technology.
The introduction of Shift & Drift provides a standardized method for rigorously testing autonomous driving AI against distribution shifts and perturbations, driving innovation towards more resilient systems.
- · Autonomous vehicle developers
- · AI safety researchers
- · Insurance companies
- · Motion planners with poor generalization
- · Companies relying on limited ID performance
- · Traditional safety testing methodologies
Motion planning AI will become more robust and capable of handling unforeseen situations.
This improved robustness could accelerate regulatory approval and public trust in autonomous driving systems.
Widespread adoption of autonomous vehicles could lead to significant shifts in urban planning, logistics, and transportation infrastructure.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG