|Títol||Learning coaching advice to improve playing skills in robocup|
|Publication Type||Journal Article|
|Year of Publication||2006|
|Authors||Bou E, Plaza E, Rodríguez-Aguilar JA|
|Journal||AAMAS 2006 Workshop on Adaptation and Learning in Autonomous Agents and Multiagent Systems|
Coaching is a way to help an agent or a group of agents to learn and/or to improve his/their performance, but learning to coach is a tough challenge. In this paper we try to make headway in this matter by presenting a learning method that uses decision trees to learn pass advices from observations of players’ actions in the simulated RoboCup soccer environment. We propose and evaluate different learning techniques to build decision trees to from which passing advices can be generated. Finally, we empirically demonstrate that exploiting learnt advices can significantly improve the number of successful passes between teammates.
- Quant a IIIA