Training a Multiagent Hive Brain for Coordinated UGV Operations (4.4.2)

Abstract : This project builds on a learning method developed previously under DARPA support called multiagent HyperNEAT that evolves a set of neural controllers for a team of collaborating wheeled robots. The project was also supplemented by DARPA CSSG Phase 3 matching grant N11AP20003 during its first year. The research focused on three key directions: The first (1) is to extend multiagent HyperNEAT to allow evolving a team of robots that can send signals to one another over wireless connections directly from neurons in one agent to neurons in another, thereby facilitating tight coordination among robots can evolve without any explicit communication language. The second direction (2) is a novel approach, called reactivity, which facilitates robust transfer from behaviors trained in simulation to robots in the real world. The third direction (3) is to add directionality to the communication system so that agents can efficiently decide and perceive from where in space signals originate. These three complementary ideas, plus enhancements to the underlying algorithms, have appeared so far in 8 conference and 4 journal articles. One paper, at IJCNN-2012, won a Best Student Paper Award out of a pool of 299 entries. Another was a Paper Award Finalist at ICIRA 2012.