Closed-Loop Dynamic Control of a Soft Manipulator Using Deep Reinforcement Learning

The focus of the research community in soft robotic field has been on developing innovative materials, but design control strategies applicable to these platforms is still an open challenge. This due their highly nonlinear dynamics which difficult model and degree stochasticity they often incorporate. Data-driven controllers based neural networks have recently explored as a viable solution be employed for manipulators. letter presents network-based closed-loop controller, trained by deep reinforcement learning algorithm called Trust Region Policy Optimization (TRPO). training takes place simulation, using approximation robot forward dynamic obtained with Long-short Term Memory (LSTM) network. controller allows following different paths executed velocities workspace robot. results demonstrate that effective normal working conditions payload attached end-effector manipulator.