Reinforcement Learning-Enabled Closed-Loop Adaptive Control Soft Robots for Biomedical Applications

Abstract

The inability of the conventional rigid robots to be flexible, compliant and safe in dynamic human situations is a hamper in the growing use of robotic technologies in the biomedical and assistive technologies. This limits their application in rehabilitation systems, prosthetics and minimally invasive surgery.Although soft robots offer an efficient and biomimetic answer, not all of them employ closed-loop smart control, which hampers their efficiency.We suggest a closed-loop adaptive control system of soft robots using electroactive polymers (EAPs), hydrogels and reinforcement learning (RL). This system is based on the principle of quick reaction of EAPs and biocompatibility of hydrogels to create effective movements. The controller is optimized using more than one sensory feedback (strain, pressure, bio-signals) and RL.Design, modeling, simulation and experimentation are used. The RL controller seeks to maximise performance measures (accuracy, speed, energy) under varying conditions.The results indicate improved flexibility, accuracy of control, and power consumption in comparison to the traditional rigid and non-intelligent soft robots. This project offers a smart adaptive system that has built-in sensing, actuation and real time control.The developed system has potential to be applied in robotic rehabilitation systems, smart prosthetics and minimally invasive surgery systems, thus promoting next-generation personalised medical robots.