Autonomous Control Of Redundant Hydraulic Manipulator using Reinforcement Learning with Action Feedback

This article describes a data-driven approach to autonomously control hydraulic manipulators with minimal system information. The hydraulic actuation dynamics are modeled using actuator networks, which emulates the real system in a simulation environment. The approach uses a neural network control policy based on end-effector position tracking learned through Reinforcement Learning (RL) with Ornstein-Uhlenbeck process noise for efficient exploration. The proposed approach is implemented on a hydraulic forwarder crane to track the desired position of the end-effector in 3D space, and the results demonstrate the feasibility of deploying the learned controller directly on the real system.