OGM-Based Real-Time Obstacle Detection and Avoidance Using a Multi-beam Forward Looking Sonar

Abstract

Autonomous underwater vehicles (AUVs) have a limited bandwidth for real-time communication, limiting rapid responses to unexpected obstacles. This study addressed how AUVs can navigate to a target without a pre-existing obstacle map by generating one in real-time and avoiding obstacles. This paper proposes using forward-looking sonar with an occupancy grid map (OGM) for real-time obstacle mapping and a potential field algorithm for avoiding obstacles. The OGM segments the map into grids, updating the obstacle probability of each cell for precise, quick mapping. The potential field algorithm attracts the AUV towards the target and uses repulsive forces from obstacles for path planning, enhancing computational efficiency in a dynamic environment. Experiments were conducted in coastal waters with obstacles to verify the real-time obstacle mapping and avoidance algorithm. Despite the high noise in sonar data, the experimental results confirmed effective obstacle mapping and avoidance. The OGM-based potential field algorithm was computationally efficient, suitable for single-board computers, and demonstrated proper parameter adjustments through two distinct scenarios. The experiments also identified some of challenges, such as dynamic changes in detection rates, propulsion bubbles, and changes in repulsive forces caused by sudden obstacles. An enhanced algorithm to address these issues is currently under development.