A Multi-Strategy Architecture for On-Line Learning of Robotic Behaviours using Qualitative Reasoning

摘要: A Multi-Strategy Architecture improves the efficiency of on-line learning robotic behaviours by taking inspiration from approaches humans use for complex behaviours. The hybrid approach first learns qualitative dynamics a system which symbolic planner constructs an approximate solution to control problem qualitatively reasoning over discov- ered dynamics. parameters are refined numerical optimization, into policy reactive controller. is demonstrated on multi-tracked robot intended urban search and rescue.