Section: New Results

AI for real time strategy games

Participants : Santiago Ontanon, Florian Richoux, Alberto Uriarte.

We continue to develop an artificial intelligence, AIUR , to play the real time strategy (RTS) game $StarCraf{t}^{tm}$, using both machine learning and constraint-based techniques. AIUR  finished 4^th over 18 finalists to the $StarCraf{t}^{tm}$ AI competition organized at the conference AIIDE 2014 , and 4^th over 13 finalists to the competition at CIG 2014 . This year, we wrote an ad-hoc csp solver to deal with the wall-in optimization problem [36] for StarCraft, and generalized it as a framework enable to handle any kind of csp /cop models representing a RTS-related problem. This framework, named GHOST , helps the user to implement his csp /cop model before solving it with the ready-to-use, already-tuned embedded solver.