Effects of System Time Delay as Humans Learn to Control Dynamic Systems

We present results from an experiment in which 22 human subjects each interact with a dynamic system 50 times over a one-week period. For each interaction, a subject is asked to perform a command-following task, which is the same task for all 50 of the subject's trials. However, the time delay of the dynamic system is increased twice during the 50 trials. We use the experimental results to examine the effects of system time delay on the subjects' performance, control strategies, and learning process. For example, we examine the effects of time delay on the subjects' step-command following performance (e.g., transient error, steady-state error, settling time). We also use subsystem identification to model the control strategies (feedback, feedforward, and feedback delay) that each subject uses on each trial of the experiment. The average identified feedforward controller approximates the inverses dynamics of the system with which the subject interacts better after numerous trials than on the first trial. In addition, increasing the system time delay tends to degrade the subjects' ability to approximate the inverse system dynamics in feedforward.