Next: Bibliography Up: Future Work Previous: Real experimentation
Besides the use of CBR described in the Genetic Algorithm approach, we also plan to add a CBR agent that would bid for actions. This agent would use the information of past experiences in different trials (stored in form of {situation,action,result} tuples) to recognize similar situations, and would then bid for executing the actions (or similar actions) that best suited those situations. The difficulty of this approach is to find the proper way to characterize the situations and how to compare two situations in order to find out how similar they are. In this approach we also face the credit assignment problem, since we cannot evaluate a situation-action experience until the robot either successfully reaches the target or fails in its mission.