Adaptive PID control of the drive system for tracked agricultural robots operating in unstructured terrain

This paper proposes a computationally efficient adaptive PID control method to enhance the drive system performance of tracked agricultural robots operating on unstructured terrain. Conventional fixed-parameter PID controllers struggle to adapt to variations in soil properties and dynamic load disturbances, while advanced control algorithms such as sliding mode control and fuzzy PID, though theoretically more adaptable, suffer from high computational complexity that hinders their practical deployment in resource-constrained embedded systems of agricultural robots. Building upon the classical PID structure, the proposed method introduces an online parameter adaptation mechanism based on Lyapunov stability theory and gradient descent, ensuring closed-loop stability and guaranteeing that the tracking error converges to a tunable residual set around zero in the sense of uniform ultimate boundedness. Comparative experiments conducted on a self-developed tracked robot platform demonstrate that, compared to fixed-parameter PID, the proposed method reduces the average RMSE of the left wheel from 0.1198 rad/s to 0.0971 rad/s (18.9% improvement) on structured surfaces, and from 0.2186 rad/s to 0.1201 rad/s (45.0% improvement) on unstructured surfaces, while the average RMSE of the right wheel decreases from 0.1231 rad/s to 0.0907 rad/s (26.3% improvement) and from 0.1925 rad/s to 0.1085 rad/s (43.6% improvement), respectively. Additionally, the control output remains smooth and chatter-free. In terms of algorithm complexity, the method achieves reductions of 23.7%, 75.8%, and 51.4% compared to sliding mode control, adaptive sliding mode control, and fuzzy PID, respectively, significantly lowering the implementation barrier in embedded systems. This approach provides a practical solution for agricultural mobile robot drive control that effectively balances control accuracy, environmental robustness, and ease of implementation.