Neural network based human performance modeling
Edward L. Fix · IEEE Conference on Aerospace and Electronics · 2002
A neural network architecture, derived from recurrent backpropagation, which learns to mimic human behavior and performance in a sample task is presented. It shows operating characteristics similar to those of human subjects, and even makes the same kinds of mistakes. The goal of this task was to develop a neural network which, when trained on data derived from a human subject performing a task, emulates that subjects's performance and style. The task was to be interactive, and the conditions controlled. The task the subjects performed was based on a popular neural network demonstration concept. It is a computer generated display showing a two-lane circular track and five cars. One car is controlled by the subject, and the other four are controlled by the computer. The cars all travel in a counter-clockwise direction. The perspective is adjusted so that the subject's car is always at the 3 o'clock position on the track, and everything else moves relative to the controlled car. The subject's task was to drive his car around the track, switching lanes and adjusting speed as necessary to avoid collisions. Results and possible applications are discussed.>