Boosting Robustness for All-Weather Self-Supervised Depth Estimation in Autonomous Driving

Researchers proposed an all-weather self-supervised depth estimation pipeline for autonomous driving.

The method uses unpaired real data, uncertainty-aware multi-teacher distillation, and POV-to-BEV radar fusion to handle degraded camera views and sparse radar inputs.

It is designed to improve robustness in rain, fog, and other adverse conditions, where reliable depth sensing is critical for safety.

The paper reports state-of-the-art performance on all-weather datasets.