Intelligent diving guidance with terminal angle and velocity constraints via deep reinforcement learning

An adaptive guidance strategy that integrates optimal guidance and deep reinforcement learning to address a highly dynamic terminal guidance problem that entails meeting terminal position, angle, and velocity constraints. The proposed strategy leverages optimal guidance commands to accomplish position and angle control while introducing a deep reinforcement learning-based bias for the velocity constraint. In the training process, a dual-velocity state space is constructed to enhance the adaptability of the strategy to different guidance tasks, while training is optimized using the prediction-correction and expert knowledge to improve the training efficiency and optimality of the strategy. Simulations demonstrate that the proposed guidance strategy can achieve simultaneous control of terminal position, angle and velocity, and adapt to different guidance tasks.