Teamwork Planning and Learning in Adversarial Multi-Robot Domains
We have been participating in the RoboCup robot soccer small-size league for more than ten years. Such multi-robot domain offers a variety of challenges, including perception, control, actuation, and teamwork. In this talk, I will present an overview of the advances we have made along these years, namely in efficient vision processing, in physics-based motion planning, in adapting to the opponent, and in coordination and teamwork. At the end, I will outline the directions of our current work, in particular in building large teams of small robots, and analyzing and learning from data logged from the robots fast execution.