Previous: A simulation study: the Mobile Robot Lab Up: A simulation study: the Mobile Robot Lab Next: The physical environment description
A major objective for this project is learning emergent behaviors. Like Agre with his improvised actions [Agre \& Chapman 1987] and Brooks with his subsumption architecture [Brooks 1985] we believe complex behaviors emerge from interaction of the agent with its environment without planning. However, previous work in this area hard-coded a lot of primitive actions. Furthermore, it did not attempt to learn the improvised behavior. In this simulation, we plan to start with a minimal number of primitive actions and sensations. Our basis for this minimality and the choice of primitive actions is physiological. In other words, in our modeling an agent, we will choose actions that are physically basic for the agent's body as primitive. We then instruct the agent to perform tasks and in the midst of accomplishing this, we expect it to notice some types of behaviors emerge. An example of an emergent behavior we will explore is moving toward an object. We expect the agent to be learning to coordinate its wheel motions, starting from nothing more than the primitive sensation of contact with an external object, and the primitive actions of turning its motors independently on or off.