Experimental evaluation of adaptive CMAC haptic control for teleoperation of compliant-joint manipulators

Abstract

Using compliant joints and haptic feedback are two methods that could improve next-generation teleoperated systems, but both methods present challenges to traditional control systems. This paper describes the first experimental investigation of a new approach to real-time haptic control for teleoperated robots with compliant joints. One original aspect of the approach is using an auxiliary error, in which a velocity penalty is added to the force error. Also, the method utilizes a Cerebellar Model Articulation Controller (CMAC), a type of neural network known for its rapid adaptation. The adaptive neural network compensates for unknown nonlinear system dynamics, interaction with unstructured environments, and non-passive operator behaviour in real-time. The auxiliary error damps vibrations and allows for control of the robot in free space without the need for control switching. In real-time experiments with both computer-generated trajectories and full bilateral teleoperation, the proposed controller tracks torque as well as a custom PID controller and outperforms the PID during free-space velocity tracking. In addition, the proposed approach causes significantly less control signal chatter than the PID during full bilateral teleoperation. A Lyapunov analysis guarantees that the proposed controller has uniformly ultimately bounded signals.