Cooperation via Indirect Communication

Evolution of indirect communication between two embodied agents is the research topic addressed in this thesis. Indirect communication is a niche that should be more common in intelligent robotics as it is essential in most areas where humans are present. Indirect communication is robust and can serve as either a supporting tool or a substitute for other means of communication. Evolutionary approach to emergence of the indirect communication will be investigated in NAO robots, as one robot will train another robot to understand the meaning of gestures given by the former. Cognitive architecture of the trainer will be evolved using HyperNEAT algorithm with a threshold to enforce modularity of the evolved controller, where every module will be responsible for a particular function of the robot. The evolved controller will be tested in reduced RoboCup environment with one robot being a goalie and a trainer and another a kicker. This report summarises available literature on the topic and displays a plan for the implementation of the system.