Action selection in robotics is a challenging task: the robot has to reason about its world beliefs, i.e. the state of the environment, and rationally act in consequence in order to complete a task (typically divided in subtasks). Moreover, in the case of a robot team, robots must agree on the decisions made (who and what to do to complete the subtasks), jointly execute the actions, and coordinate among them to successfully perform the task. Working with real robots has additional difficulties that must be considered. Thus, the reasoning engine must be capable of dealing with high uncertainty in the robot?s perception (incoming information of the world), and be robust in case of failure, since the outcomes of the actions performed are unpredictable. Not to mention that decision must be made on real time and in our case, with limited computational resources. Within this framework we introduce a problem domain which fulfills most of the characteristics presented above: robot soccer. In this domain we face a dynamic and unpredictable environment where a team of robots select the actions to perform in order to reach its goal: to win a game. More precisely, we focus our work on the Four-Legged League, one of the leagues in the RoboCup competition. We propose a Case-Based Reasoning approach to solve this problem of action selection in the robot soccer domain.
- Acerca del IIIA
- 25 aniversario