Overview and Practical Evaluation of Sonar-Based Collision Avoidance Approaches for AUVs

One of the main challenges in the field of autonomous underwater vehicles (AUVs) is the collision avoidance. Under the aspect of low-weight and low-cost AUVs, the computation time and the capabilities of the given sensors e.g. forward looking sonar (FLS) are limited in terms of data quality and functionality. In this paper a comprehensive summary of current collision avoidance methods is presented and specific ones are evaluated. The evaluation is performed for the different stages of the collision avoidance: The pre-processing of FLS data, the object recognition and collision detection, and the path re-planning for scheduling an alternative and secure path. The evaluation is performed using real sonar data as well as simulations. The most promising approach of each stage is implemented and examined in a field trial.