Robust Underwater Obstacle Detection for Collision Avoidance

Abstract

Underwater obstacle detection and avoidance is essential for safe deployment of autonomous underwater vehicles (AUVs). A forward-looking sonar is typically used to detect and localize potential obstacles. Such sensors tend to have a coarser sensor resolution and a lower signal-to-noise ratio (SNR) than electromagnetic sensors typically used for similar tasks in land-based robotics. Lack of access to GPS causes additional uncertainty in vehicle navigation, making it difficult to detect and localize potential obstacles relative to a world-fixed reference frame. In this paper, we propose an obstacle detection algorithm for AUVs which is based on occupancy grids. The proposed method differs from existing occupancy grid-techniques in two key aspects. First, we use an occupancy grid attached to the body frame of the AUV, and not to the world frame. Second, our technique takes detection probabilities and false alarm rates into account, in order to deal with the high amounts of noise present in the sonar data. The proposed algorithm is tested online during field trials at Pandan Reservoir in Singapore and in the sea at Selat Pauh off the coast of Singapore.