Benchmark RadarShift & Drift
Shift & Drift evaluates closed-loop motion planners on two tracks: Semantic Shift, which uses a conversion pipeline to transform the DeepScenario Open 3D dataset into nuPlan for zero-shot testing on 1,182 scenarios across German cities and San Francisco, and State-Distribution Drift, which injects stochastic perturbations into ego-vehicle dynamics. Scoring is based on safety and progress metrics.
- Released
- 2026-07-08
- Readiness
- Paper only
- Primary field
- Transport & Logistics
Why it matters
Addresses the evaluation gap in generalization of motion planners to novel urban topologies and robustness to execution perturbations, providing a dual-track benchmark that quantifies the trade-off between imitation fidelity and closed-loop resilience.
Motivation
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.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.