Unified Risk Map Framework Combines Traffic Flow and Collision Risk Under Occlusion

The framework integrates traffic flow risk and collision risk via spatiotemporal modeling for partially observable driving environments, and uses a diffusion-based scenario generator to produce adversarial occlusion scenarios for training data augmentation. It targets the gap between overly conservative reachability-based methods and trajectory predictors that fail under high occlusion uncertainty.