SenseWeather: Multimodal Sensing for Robust Robotic Perception under Extreme Conditions

This research introduces SenseWeather, a controlled benchmark for robotic perception under extreme weather conditions such as rain, snow, and fog. The paper proposes Fusion-to-RGB techniques using sensor fusion to generate task-compatible RGB observations, improving downstream perception tasks without retraining models. Results show that multimodal sensing enhances perception robustness, with improvements in detection performance across various models.