Robust Perception Under Domain Shift
Watch camera, LiDAR and radar confidence diverge as weather severity rises, then compare naive averaging against uncertainty-weighted sensor fusion and see when out-of-distribution detection should trigger a safe fallback.
Self-driving cars and mobile robots fuse camera, LiDAR and radar readings into a single confidence estimate — but each sensor degrades differently once weather pushes the input outside its training distribution. This simulator drives that gap: dial up fog, rain or snow severity and watch camera, LiDAR and radar confidence diverge, then compare naive averaging against uncertainty-weighted fusion to see why weighting by per-sensor reliability keeps the fused estimate — and an out-of-distribution alert that hands off to a safe fallback — trustworthy even when one modality is blinded.
Dial up fog, rain or snow severity and watch camera, LiDAR and radar confidence diverge as domain shift sets in, then compare naive averaging against uncertainty-weighted sensor fusion and see when out-of-distribution disagreement should trigger a safe fallback.
3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install