AUV Adaptive PID Control Method Based on Deep Reinforcement Learning

Abstract

Intelligent control systems have evolved to manage motion systems with intricate dynamics. Despite this, the straightforward nature of PID (Proportional Integral Differential) controllers keeps them in prevalent use in industries and robotics. This paper introduces an intelligent control mechanism utilizing DRL for the adaptive tuning of multiple PID controllers in autonomous underwater vehicles (AUVs). Our innovative approach incorporates a hybrid control scheme with a participant critic framework, processing only essential low-level dynamic data while concurrently adjusting various parameters or gains within the PID controllers. Extensive simulation testing has demonstrated the practicality of this approach for basic AUV control. Both simulation and real-world experiments indicate that our method effectively utilizes behavioral adjustments to counterbalance or adapt to uncertainties in the environment. thus helping to provide unsupervised solutions without models. In addition, it is compared with other adaptive methods for multi PID adjustment, and shows the successful performance of this method.