Long Road Ahead: Lessons Learned from the (soon to be) Longest Running Cognitive Model

We present a cognitive model that plays a video game of driving a bus for a long time. The model was built using the ACT-R cognitive architecture and an extension to support perceptual-motor knowledge of how to interact with the environment (VisiTor and ACT-R/PM). Our extension includes bitmap-level eyes and robot hands. The model was run for a long time, over 6 hours on the way from Tucson to Las Vegas. We employed a design approach based on the ADDIE model to create different knowledge representations and actions; the model’s predictions can be matched to some aspects of human behavior on the fine details regarding the number of course corrections and average speed and learning rate. However, it does not exhibit the same level of fatigue as human behavior. This contrasts with the way humans typically perform such long tasks. This model shows that perception opens up new interfaces and provides a very accessible testbed for examing further aspects of behavior. and adding components of human behavior that remain missing from ACT-R.