Reinforcement Learning-Based Online Drive Control for MEMS Gyroscopes

Micro-electromechanical systems (MEMS) gyroscopes are highly sensitive to environmental disturbances such as time-varying temperature, vibration, and random noise. To address these issues, this work proposes an online intelligent control method based on the reinforcement learning, implemented using the proximal policy optimization (PPO) algorithm. A gyroscope dynamical model with error terms is first established, followed by the formulation of the control problem as a Markov decision process. An online control framework is then developed, in which control actions are adaptively adjusted according to the observed system states. A hardware-in-the-loop (HIL) experimental platform is constructed to enable the real-time control of a physical gyroscope. Experimental results show that, during the start-up phase, the proposed method reduces the drive-voltage overshoot by 90% compared with a PID controller. During long-term steady operation, the bias stability is improved by 26%. These results demonstrate that reinforcement learning (RL)-based online control can enhance both the transient response and long-term stability of MEMS gyroscopes under real-world disturbances.