Global Sliding Mode Guidance Law with Intersection Angle Constraint Based on Reinforcement Learning

A global sliding mode guidance law (GSMG) incorporating reinforcement learning (RL) is proposed to handle guidance tasks with intersection angle constraints. Firstly, the connection between the desired intersection angle and the line-of-sight (LOS) angle is established. A GSMG law is constructed, ensuring system stability, with an adjustable coefficient introduced for further refinement. Secondly, RL is leveraged to optimize this coefficient while reducing the number of observation variables. A deep deterministic policy gradient (DDPG) algorithm is employed for training, with a specifically designed network structure and reward function. Thirdly, the agent learns to output optimized coefficients, and comparative simulations validate the effectiveness of the guidance strategy.