ABSTRACT
In this chapter, we present a gliding efficiency optimization strategy based on deep reinforcement learning (DRL) and a separate strategy to control the pitch attitude and the gliding direction for a bionic gliding underwater robot. First, a concept of transient gliding motion is presented, and several pectoral fins with different sizes are designed and analyzed by computational fluid dynamics simulation. A double-deep Q-network–based optimization strategy is proposed to improve gliding efficiency by active pectoral fins. Second, aiming at the separate control of the pitch attitude and gliding direction, the gliding dynamic model is decomposed into pitch and velocity terms. The backstepping and model predictive controllers are designed to regulate pitch angle using a movable mass and angle of attack using pectoral fins based on their control features, respectively. Simulations are conducted to validate the effectiveness of the proposed strategies, including the network training and control of the gliding optimization strategy, as well as the motion control and tracking of the separate control strategy. Furthermore, to capture the real-time gliding states in a practical environment, the authors develop a gliding measurement and control system, through which the aquatic experiments are carried out to further verify the strategies. The results reveal that the optimization strategy can save about 4.88% of energy and 19.45% of travel time while a separate control strategy is effective for both single and concurrent control of the pitch attitude and the gliding direction.
