REACTIVE, SEQUENCED Q-LEARNING OF MULTIPLE TASKS BY AN AUTONOMOUS MOBILE ROBOT
暂无分享,去创建一个
Literature review. Different ways researchers have handled multiple tasks: 1) Separate learning of tasks (sub-tasks) and design an arbiter to select which Q-table to follow [Mahadevan & Connel 2) Learning for Easy Missions (LEM). Decompose a goal oriented task: from easy situations to subsequently more and more difficult ones. Addresses the problem of infrequent and delayed reinforcement. [Asada et al. 1996], 3) Decomposition of tasks: a) Switching between task by means of an arbiter. [Whitehead et al. 1993]
[1] Jonas Karlsson,et al. Learning Multiple Goal Behavior via Task Decomposition and Dynamic Policy Merging , 1993 .
[2] Leslie Pack Kaelbling,et al. Learning in embedded systems , 1993 .
[3] Maja J. Mataric,et al. Interaction and intelligent behavior , 1994 .
[4] Sridhar Mahadevan,et al. Introduction to Robot Learning , 1993 .