A Study on Underwater Autonomous Vehicle Terrain-following Depth Control Algorithm using Reinforcement Learning Method

In this paper, we designed a depth controller to enable an underwater autonomous vehicle to follow depth commands while avoiding the seafloor topography. A linear dynamic model was developed and applied, and reinforcement learning based on the PPO algorithm was utilized for depth command generation. This method leads to improved flexibility and high accuracy of depth command. For verification of algorithm performance, driving performance was analyzed for a scenario that applied actual underwater topography data for around the Korean Peninsula. The proposed algorithm is expected to be used to establish a route for maximizing driving efficiency through terrain avoidance and tracking for underwater autonomous vehicles.