Obstacle Perception

Obstacle perception, crucial for autonomous systems, aims to accurately identify and classify obstacles in the environment to enable safe navigation. Current research emphasizes multimodal sensor fusion, particularly combining cameras (including fisheye) and ultrasonic sensors, often employing deep learning architectures like ResNeXt and novel fusion strategies to improve robustness and accuracy, especially in challenging conditions. This work also focuses on refining safety zones around obstacles using maneuver-based decomposition and developing computationally efficient methods for resource-constrained platforms like robots and UAVs, improving the reliability and practicality of autonomous navigation in diverse environments.