An Improved DDPG Reinforcement Learning Control of Underwater Gliders for Energy Optimization

摘要

As a novel underw ater vehicle, underw ater gliders are widely used in marine environment exploration. Underwater gliders are designed for long-term and longdistance operation, adaptivity and energy optimization is a critical requirement for controller design. In this paper, the reinforcement learning control is studied for underwater gliders, and the problem of slow learning convergence and unstable learning process of the DDPG reinforcement learning algorithm. The proposed solution is based on the priority experience replay method, which effectively increase the convergence speed and stability of the algorithm is addressed. The gliding control parameters are optimized to reduce the energy consumption is proposed, by using the improved DDPG algorithm and the energy consumption model. In the simulation experiments with an underwater glider, a set of glide parameters is obtained at a given gliding depth.