Autonomous Control Of Redundant Hydraul...

This article presents an entirely data-drivenapproach for autonomous control of redundant manipulatorswith hydraulic actuation. The approach only requires minimalsystem information, which is inherited from a simulation model.The non-linear hydraulic actuation dynamics are modeled usingactuator networks from the data gathered during the manualoperation of the manipulator to effectively emulate the realsystem in a simulation environment. A neural network controlpolicy for autonomous control, based on end-effector (EE)position tracking is then learned using Reinforcement Learning(RL) with Ornstein–Uhlenbeck process noise (OUNoise) forefficient exploration. The RL agent also receives feedbackbased on supervised learning of the forward kinematics whichfacilitates selecting the best suitable action from exploration.The control policy directly provides the joint variables asoutputs based on provided target EE position while takinginto account the system dynamics. The joint variables are thenmapped to the hydraulic valve commands, which are thenfed to the system without further modifications. The proposedapproach is implemented on a scaled hydraulic forwarder cranewith three revolute and one prismatic joint to track the desiredposition of the EE in 3-Dimensional (3D) space. With theemulated dynamics and extensive learning in simulation, theresults demonstrate the feasibility of deploying the learnedcontroller directly on the real system.