RESBev: Making BEV Perception More Robust
Researchers have developed RESBev, a new method to enhance the robustness of Bird's-Eye-View (BEV) perception systems used in autonomous driving. This plug-and-play technique can be integrated with existing BEV models to improve their resilience against sensor degradation and adversarial attacks. RESBev works by predicting clean BEV features from corrupted observations using a latent world model that captures spatiotemporal correlations. Experiments on the nuScenes dataset show significant improvements in robustness with minimal fine-tuning. AI
IMPACT Enhances the safety and reliability of autonomous driving systems by making perception more resilient to real-world disturbances.