Research

Our scientists are at the forefront of developing reinforcement learning techniques to tackle challenges involving multiple cooperative, competitive, or hybrid machines. Our research formalized the centralized planning for the decentralized control paradigm, an elegant framework for reasoning in artificial intelligence systems with imperfect information that underpins almost all multi-agent reinforcement learning algorithms. See  ANR Projects: PLASMA (closed), DELICIO  (ongoing), MAESTRIOT  (ongoing), EpiRL  (ongoing).