Reinforced Learning based Resonant Frequency Tracking for Percussive Ultrasonic Drills

In situ exploration of the probe on the asteroid surface requires anchoring mechanisms to overcome the difficulties of weak gravity, extreme temperatures, and unknown landscapes. The percussive ultrasonic drill is a drill tool that requires low drilling pressure, a wide range of temperature and high drilling capacity, making it suitable for anchoring tasks on asteroid surfaces. In order to maintain efficient and long-lasting drilling of percussive ultrasonic drill and solve the problems of random noise interference and resonant frequency drift during drilling control, a reinforcement learning resonant frequency tracking control strategy is proposed. We investigate the way of training and updating the frequency step action with the phase angle absolute waveform as the tracking object, and design the method of frequency step change that closely follows the frequency drift step. It is experimentally demonstrated that compared with fuzzy control and SARSA, the proposed reinforcement learning control strategy is able to effectively track the resonant frequency of percussive ultrasonic drills that are disturbed by random fluctuations and drifted nonlinearly, and drive the ultrasonic drills faster into different rocks.