UCLA

Oscillatory movements play important roles in human life, various movements in human life, such as walking, bicycling, cleaning, chewing, swimming, etc., are periodic or repetitive. A robotic device that can assist human with such motion tasks would be of utterly significant. This research work addresses the control design problem for such robotic devices and investigates the methods for designing feedback controllers for a robotic system to help a human with periodic motion tasks. The control objective is to stabilize a human-intended oscillatory movement while reducing the required human effort.

To approach this problem, two mathematical models are studied and formed as controllers accordingly for a general mechanical system. First, a biological neural circuit model, central pattern generator (CPG), is adopted. Animal locomotions under CPG control are capable of complying with various environment dynamics to yield different oscillatory movements. We take advantage of a mathematical model of reciprocal inhibition oscillator (RIO), a simple-structured and well-studied type of CPG. The RIO controller acts as a nonlinear damping compensator and removes part of the resistive forces in the system, thereby reducing the human effort. It is shown that the resulting human-intended oscillation is a locally stable periodic solution of the closed-loop system, assuming a simple human intention motor control. The result is first presented for a single degree-of-freedom (DOF) mechanical system and then extended to a multi-DOF system.

Alternatively, a nonlinear oscillator model, Andronov-Hopf Oscillator (AHO), is selected. Nonlinear oscillators are appropriate candidates for controlling such systems since they are capable of generating stable rhythm signals. Here we consider the problem of designing a nonlinear adaptive feedback controller for uncertain linear mechanical systems so that convergence to a natural mode of oscillations is achieved for the closed-loop system. We propose a controller based on the Andronov-Hopf oscillator with additional adaptation mechanisms for estimating the unknown natural frequency and damping parameters. We prove that, with sufficiently slow adaptation, the estimated parameters locally converge to their true values and entrainment to the natural oscillation is achieved as part of an orbitally stable limit cycle. Numerical examples demonstrate that adaptation and convergence can in fact be fast.

To examine the research results, a four-linkage robotic arm system is designed and prototyped. With the hardware set-up, the proposed human intention motor control is validated by first identifying the control parameters and then replacing human to control the robotic arm. Furthermore, it is experimentally shown that robotic arm under RIO control is able to stabilize human-intended oscillatory movements and reduce the human eeffort.