Extended Case-Based Behavior Control for Multi-Humanoid Robots

Abstract

Intelligent control of multi-agent autonomous humanoid robots is a very complex problem, especially in the RoboCup domain. This is due to the dynamics of the environment and the complexity of behaviors that should be executed in real time. Moreover, the overall complexity increases since another mechanism for coordinating the robot’s behaviors should be involved. Throughout the past years, acceptable results have been obtained using different approaches to solve the behavior control problem (decision trees, fuzzy techniques, support vector machines, reinforcement learning, etc.). As case-based reasoning (CBR) is tightly related to the way humans reason, researches are now concentrated on finding the most relevant solutions that could handle behavior control problems of humanoid soccer robots using CBR techniques. This paper proposes an extended algorithm of a more complex architecture to control more complex soccer behaviors such as dribbling and goal scoring applied to multi-robot scenarios between attacker and goalie. The decomposition of features into a hierarchy of levels has led to reduced complexity of the overall behavior execution in real time. However, the average retrieval accuracy and the average performance accuracy were relatively low (70.3% and 77%, respectively). We intend to combine neural networks with CBR to learn adaptation rules for reducing the complexity of hand-coded rules and improving the overall controller performance.