Deep reinforcement learning-based attitude control for high-speed vehicle in dive phase

Aiming at the control problems of the high-speed vehicle in dive phase, such as drastic parameter changes, large uncertainty and high-precision attitude tracking requirements, an Enhanced Twin Delayed Deep Deterministic Policy Gradient intelligent controller (ETD3) is proposed to perform end-to-end control of the three-channel attitude angle, and its adaptability is verified under 13 simulation deviations. The simulation results show that the proposed intelligent controller can accurately track the three-channel attitude commands with a steady-state tracking error of less than 0.84° over a wide range of deviations in the model and environment.