Behavior Coordination for a Mobile Visuo-Motor System in an Augmented Real-World Environment
Dimitrij Surmeli, Horst–Michael Groß · The MIT Press eBooks · 2002
We utilize Humphrys ’ W-Learning on a real robot Khepera to coordinate three behaviors in an augmented maze: first, to drive straight and fast while avoiding obstacles, second, to find a lo-cation marked by one projected color (e.g., where food can be found), and third, to escape from an-other color. We describe the experimental setup and compare results of the individual agents to those of a monolithic agent solving all tasks, and of the agents coordinated by different types of W-Learning. We demonstrate the feasibility of W-Learning on a real visuo-motor system and con-clude by discussing why the monolith outperforms all forms of coordination investigated. 1