Nonlinear and reinforcement learning control for motion of hybrid aerial underwater vehicle

Hybrid aerial underwater vehicle (HAUV), a new type of vehicle that can operate in the air and underwater, has emerged in recent years and facilitates the joint aerial and underwater missions. Due to the large differences between air and water medium environments, HAUV is a complex system and is challenging to control, especially in the air/water transition. This paper proposes the nonlinear control and deep reinforcement learning control of fixed-wing HAUV (FHAUV). First, the three-dimensional space motion model of FHAUV is developed, and three key issues are involved: disturbance and uncertainty, the air/water transition, control input dead zone and saturation due to the medium differences. Then, for FHAUV and the issues, a nonlinear control including robustness, adaptation and fuzzy logic is proposed, the air/water transition is visualized, and a deep reinforcement learning control of FHAUV is designed by deterministic policy, neural network and temporal difference learning. Finally, the effectiveness of nonlinear control and deep reinforcement learning control of FHAUV is verified in the trajectory tracking control of three key issues. The control results and performance indexes are compared and analyzed. Two methods achieve the goal, the nonlinear control of FHAUV has faster response, and the deep reinforcement learning control of FHAUV has better convergence time and overall error.