MULTI-PLAN RETRIEVAL AND ADAPTATION IN AN EXPERIENCE-BASED AGENT
Ashwin Ram · 1996
To address these issues, we propose a theory of experience-based agency which specifies how an agent with the ability to richly represent and store its experiences could remember those experiences with a context-sensitive, asynchronous memory, incorporate the relevant portions of those experiences into its reasoning on demand w ith integration mechanisms, and direct memory and reasoning through the use of a utility-based control mechanism. We have implemented this theory in the NICOLE multistrategy reasoning system and are currently using it to explore the problem of merging multiple past planning experiences. N ICOLE’S control system allows memory and reasoning to proceed in parallel; the asyn chronous and context-sensitive nature of that memory system allows NICOLE to return a “best guess” retrie val immediately and then to update that retrieval whenever new cues become available. But solving the problem of merging planning experiences requires more than just memory and control. We need mechanisms to integrate new retrieved plans into the planner’s current reasoning context whenever they are found; to en sure that those new retr ieved plans are useful, we need ways to use the planner’s current context to generate new cues that can help guide the memory system’s search. To solve these subproblems, we have developed the Multi-Plan Adaptor (MPA) algorithm, a novel method for merging partial-order plans in the context of casebased least-commitment planning. MPA allows the merging of arbitrary numbers of plans at any point during the adaptation process; it achieves this by dynamically extracting relevant case subparts and splicing those subpart s into partially completed plan s. MPA can also help guide retrieval by extracting intermediate goal statements from partial plans. In this chapter, we briefly review the properties of the real world that present challenges for the design of intelligent agents, examining in particular the need to combine past planning experiences. We then review our theory of experience-based agency and its implementation in the NICOLE system. We present the MPA algorithm, illustrate how it supports the merging of plans at any point during the adaptation pro cess, and describe its foundations in leastcommitment case-based planning. We then discuss how MPA can be integrated into various