Trajectory Tracking Control of an Agricultural Mobile Robot Based on PPO Algorithm

In the unstructured agricultural environment, the four-wheel-driven agricultural mobile robot often encounters problems such as large lateral errors and poor continuity of steering actions during trajectory tracking control on complex paths with sharp turns and high curvatures. To address these issues, this paper proposes a trajectory tracking control method that combines the Proximal Policy Optimization (PPO) algorithm with a variable single preview point. The kinematics of the agricultural mobile robot platform were modeled, and a trajectory tracking controller was designed with a PID-based longitudinal control and a PPO-based lateral control at its core. Simulations were conducted in the Mujoco environment, where complex paths with multiple curvatures and sharp turns were set up. The performance of the PPO, DDPG, and Stanley algorithms in trajectory tracking tasks was compared. The experimental results show that under the single preview point condition, the PPO algorithm performs best in handling continuous action spaces and achieving higher trajectory tracking accuracy.