Probabilistic Logic Based Reinforcement Learning of Simple Embodied Behaviors in a 3D Simulation World

摘要：

We describe here the use of an integrative AI architecture to perform reinforcement learning of simple behaviors in the context of controlling a humanoid agent in a 3D simulation world. The AI architecture, the Novamente AI Engine, is extremely flexible, incorporating a variety of carefully intercoordinated learning processes; but the work described here relies primarily on the integration of probabilistic inference with statistical pattern mining based perception and functional program execution based agent control. The bulk of the paper describes how integrative intelligence is used to enable the system to learn to play the game of “fetch in an embodied, simulation world context.

关键词： AI architecture 3D simulation word play fetch

作者: Ben Goertzel Ari Heljakka Welter Silva Cassio Pennachin Andre Senna Izabela Goertzel Teemu Keinonen Matthew Iklé Sanjay Padmane

作者单位: Novamente LLC, 1405 Bernerd Place, Rockville, MD, 20851, USA Adams State College, Alamosa, CO, 81102 USA, and with Novamente LLC

会议类型: 国际会议

会议名称: The Second International Symposium on Intelligence Computation and Applications(ISICA 2007)(第二届智能计算及其应用国际会议)

会议地点: 武汉

会议语种:英文

在线出版日期: 2007-09-21（万方平台首次上网日期，不代表论文的发表时间）