Reliable autonomous driving requires full-scene perception that couples foreground objects with dense semantic layouts. Recently, 4D millimeter-wave radar has emerged as a robust and affordable sensor, yet its sparse returns necessitate radar-camera fusion for comprehensive scene understanding. Existing radar-camera methods primarily focus on optimizing detection, while dual-task systems often have limited interaction when decoding boxes and occupancy.
To address this gap and advance radar-based multi-task learning, we propose \method, a 4D radar-camera framework for 360$^{\circ}$ full-scene perception, which models semantic occupancy as a persistent scene state rather than a terminal output.
\method{} follows a cross-modal state reasoning paradigm, where the occupancy state is modeled and propagated through stages for coarse-to-fine feature aggregation. Specifically, State-guided BEV Enhancement (SBE) strengthens intra-frame BEV representation, while Doppler-guided Temporal Fusion (DTF) preserves state evidence over longer temporal horizons.
Additionally, we extend ManTruckScenes with satellite-map-based generated occupancy labels and pair it with OmniHD-Scenes in a unified cross-dataset detection-and-occupancy protocol. The resulting experiments cover accuracy, robustness, ablation, and efficiency under one radar-camera multi-task evaluation framework. Code and labels will be released upon acceptance.
Blogger's Review: This research significantly enhances multi-task learning capabilities in radar-camera fusion by introducing a state reasoning mechanism, pushing the boundaries of autonomous driving technology. Particularly noteworthy is the depth and breadth of scene understanding achieved through cross-modal state modeling, making it a valuable contribution to the field.