Receding-horizon sampling-based sensor-driven coverage planning strategy for AUV seabed inspections

Autonomous Underwater Vehicles (AUVs) are commonly used to autonomously explore or cover an entire area with a sensor to map the area or look for objects of interest using Automatic Target Recognition (ATR) algorithms. AUV surveys are typically pre-planned by operators, often using lawnmower or zig-zag paths, and are often conducted using acoustic sensors, such as Forward-Looking Sonars (FLS) and Side Scan Sonars (SSS). The performance of such acoustic sensors depends on several environmental, target, and sensor characteristics, which are difficult to predict, making it challenging to design a good path for the AUV tasks. Sensor-driven path planning algorithms emerged as an important tool to overcome such limitations. These solutions can avoid to perform several mission attempts, and thus reducing the time required for seabed inspections and explorations. The here presented work aims to investigate an AUV tailored sensor-driven path planning solution for seabed inspections using an FLS sensor, but suitable for any acoustic or optical sensor. A two-level planning layer system is presented where the well-known RRT∗ is used as a start-to-goal planner, while in the high-level planner, the possible next-best viewpoints are generated in an RRT fashion. Random trees are grown considering the Dubins vehicle kinematic constraints, where each node is evaluated according to the expected information gain. Besides, a rewiring strategy for coverage applications to find for each viewpoint the best parent is proposed. Simulation results demonstrate the effectiveness of the proposed solution.