Reference: Morignot, P. Embedded Planning. Chicago, IL, 1994.
Abstract: We present a planner that generates nonlinear plans in real-time to control an autonomous agent. Requirements of plan generation in an embedded system are obtained by extending the action representation of [Chapman 87] and [McAllester & Rosenblitt 91]. To benefit from other dedicated capabilities of the agent, locations are represented using intended actions; Physical characteristics of an agent are represented by an arithmetic equations maintainance capability. The planner exhibits attractive real-time performances when additional knowledge of the environment is available (e.g., variables' domain). An autonomous agent embodying plan generation and monitoring has been successfully implemented. It has been tested on examples from various domains (including the ones of this document). It actually controls a mobile robot at the KSL to assist in office work.
Full paper available as ps.