Event

Release of off-road dynamics learning method SpatialBEV

1 reports · 1 sources

Micah Nye et al. proposed SpatialBEV, an off-road vehicle dynamics prediction method that queries bird's-eye-view (BEV) features around the predicted state at each step, conditioning the next-step prediction on current geometric and semantic cues to characterize terrain-vehicle interaction. The paper was accepted to CoRL 2026. The evaluation results released by the authors show that on a dataset collected across seasons, SpatialBEV reduced prediction error by 17.1% compared with physics-based and learning-based baselines; in hardware tests on high-speed (≥10 m/s) and geometrically challenging routes, the maximum lateral trajectory error was reduced by 25.3%.