Planning and state evaluation in real-time decision domains

点击翻译按钮获取中文摘要

Artificial intelligence (AI) research applied to games has a long tradition thatreaches back at least 70 years with Alan Turing's work on computer chess. Theadvantage of studying AI algorithms in this area is that games are preciselydefined, relatively small when compared to real-world decision domains, and yetsufficiently complex to pose tough research problems whose solutions can help uscreating machines of human-level intelligence. AI research has had its successesin games like chess, Othello, backgammon, and checkers - where machines now playon par with or better than the best human players. However, big challenges remain- most notably in the ancient Asian board game of Go for which to date no machinecan match the strength of good beginners. Likewise, video game AI systems arelacking learning, planning, and reasoning capabilities which results in weakperformance compared to human players.  The objective of this project is to improve the decision quality of AI systemsin real-time decision domains by means of abstractions, look-ahead search, andmachine learning of search heuristics. Our application focus will be on populargames in which beginners still outperform the best programs. The results of thisproject will increase our understanding of fundamental AI problems and will haveconsiderable impact on the real-time control domain in general and the computergames industry in particular, which is in need of credible computer controlledagents.