Human Operator Dynamics

The overall goal of this research is to determine the joint mechanical properties of the freely moving human arm, by utilizing a novel airjet perturbation device and developing time-varying nonlinear system identification methods. Applications are being pursued for clinical evaluation of motor deficits and rehabilitation, and for incorporation of human operator dynamics in teleoperator control systems.

Airjet Design

The overall goal is to study natural, unconstrained arm movements. To allow the arm to move freely, it was necessary to design a portable perturbation device. Our airjet system is a pneumatic thruster attached to the wrist, which produces high-frequency stochastic perturbations with a flat response of 75 Hz and a thrust of 4 N.

A single-axis thruster was designed first and used for experimentation with the elbow joint (Xu, Hunter, Hollerbach, and Bennet, 1991).
A computational fluid dynamic analysis has been performed to optimize the design.
A two-axis thruster has been recently designed which will allow the shoulder as well as elbow to be characterized. Tests on an artificial 1D elbow joint show that the mechanical impedance is well identified, after corrections for force sensor ringing and for acceleration loading on the force sensor by the airjet (Xu and Hollerbach, 1995).
A passive two-joint planar manipulator, with inertia, damping and stiffness attributes of the human arm, has been designed to test the two-axis thruster and multi-joint identification methods.

This device places us in a unique position to study unconstrained arm motions.

Time-Varying Nonlinear System Identification

Considerable effort has been placed on developing two robust identification methods:

a single trial method, based on exponentially weighted least squares (EWLS), and
a new ensemble method, the highpass mean removal ensemble data (HMRED) method.

These methods address problems and artifacts potentially due to ensemble methods.

The EWLS method avoids such problems by requiring only a single trial; it can capture linear time-varying dynamics in posture and slow movement (Xu and Hollerbach, 1996b).
The HMRED method is suitable for both slow and fast movements; it avoids the problem of mean removal with previous ensemble methods (Xu and Hollerbach, 1996a).

The EWLS and HMRED methods have been calibrated against each other in a slow movement task: they both yield very close results (Xu and Hollerbach, 1993). Such a calibration is particularly important because the result on the human limb identification often varies with the method used in the literature.

We have also done preliminary experiments to check whether small random force perturbations alter the mechanical properties of the elbow joint in posture. We compared the pulse responses evoked by a pulse force with or without superimposed random force perturbations. We found that random force perturbations have little effect on the pulse response, and the pulse response is consistent with the mechanical properties obtained through random perturbations (Xu and Hollerbach, 1994).

In the future, we will begin experiments on two-joint posture and motion, including the shoulder as well as the elbow.

John M. Hollerbach Biorobotics Laboratory