Human Control Strategies in Manual Pursuit Tracking of Sinusoidal Signals

Abstract:

A more recent thrust of our work in this project focuses on continuous control tasks performed by a human operator. Tracking random-appearing and oscillatory signals is a human in the loop task that has been used in many areas such as piloting of vehicles, rehabilitation engineering, and neuroscience. Understanding the control strategies in a human operator for these tracking tasks is of great importance in these areas. The major goal of this project is to develop identification methods to simultaneously estimate the feedback and feedforward controllers in a human operator. Human operators can manipulate an integrator plant and manually track a single sine wave very well with almost unity amplitude ratio and zero phase lag as long as the frequency of the sine wave is not too high. However, the pattern of responses across single sine signals of different frequencies does not resemble the pattern of response across random appearing sum-of-sines signals for moderate to high frequencies. There is a significantly larger phase lag in tracking a sum-of-sines than tracking a single sine wave for moderate to high frequencies. This phenomenon suggests that humans utilize different control strategies for predictable tasks than for unpredictable tasks. In this project we designed experiments and methodologies to identify non-parametric feedback and feedforward controllers in human operators for manual tracking with an explicit display of the sinusoidal reference signals and a disturbance input. Our results show that the feedback controller resembles McRuer's "crossover model", and the feedforward controller attempts to invert the system dynamics that the human operator is manipulating if the reference signal is predictable. If the reference signal is random-appearing and not predictable, the feedforward controller of the human operator loses the ability to invert the system dynamics. There are only two identification methods available in the literature to simultaneously estimate feedback and feedforward control in human operators. One of these, by McRuer and associates dating from the 1960s and 1970s assumes that the feedback controller does not change when supplemented by feedfoward control. The other body of work is currently underway in Max Mulder's group in Delft, The Netherlands, and assumes a parametric model structure. Compared with these existing identification algorithms, our proposed algorithm does not rely on the assumption in McRuer's early work and also does not assume any model structure. The control-oriented models of human operators in this work have applications in several fields involving a human in the loop. Examples include human skill transfer, assist for neurodegenerative disorders, and rehabilitation after neurological injury.