Sonar Mapping and Obstacle Avoidance for Autonomous Underwater Vehicles in Unknown Marine Environments

Abstract

This letter proposes a safe navigation framework for autonomous underwater vehicles (AUVs) that integrates sonar mapping and motion planning operating in unknown environments. The challenge is that sparse sonar data causes incomplete obstacle boundaries, posing risks to the vehicle's safe navigation. Furthermore, the AUV's nonlinear dynamics and constrained onboard computing capacity complicate real-time planning. To address these issues, Bayesian kernel inference method is first adopted to estimate occupancy in unobserved regions. Its accuracy is enhanced by an entropy-based adaptive approach that adjust the training set and kernel length. Then, a multilayer motion planner is designed, integrating global path planning with local trajectory generation. Specifically, a novel 3D path planner capable of handling AUV pitch and yaw constraints for efficient search is proposed, followed by path refinement and spatiotemporal optimization within the sonar's perception range to obtain locally optimal trajectories. The sonar mapping and obstacle avoidance motion planning are evaluated through simulations and pool experiments using a real AUV. Results demonstrate the effectiveness of the proposed underwater mapping and planning framework.