Case-based planning with a high-performance parallel memory

Abstract

Case-based planning (CBP) systems, like other case-based reasoning\nsystems, can take advantage of previous planning experience by reusing\nstored cases (plans) in similar situations in the future. Advantages\nof CBP include speedup over planning from scratch and the ability to\nfunction with limited causal domain knowledge. ``Traditional'' CBP\nsystems with the latter advantage typically cannot produce plans from\nscratch because they lack the more powerful adaptation mechanisms of\ngenerative planning systems. These ``reuse-only'' CBP systems rely on\nretrieving a plan from the casebase that is close to a solution plan.\nThis requires large casebases with good coverage of the problem space\nand the ability to encode and match cases at fine levels of detail.\n\nMany such CBP systems, however, have fallen short of these\nrequirements. They support only small, pre-indexed casebases.\nPre-indexing constrains retrieval, as does the use of less expressive\nfeature-based case representation schemes. The encoding and matching\nof detailed structural relationships in cases is not possible in such\nsystems. These systems often adapt a single plan to the target\nproblem using methods that are ad hoc or heuristic.\n\nCAPER is a novel, domain-independent, case-based planning system with\nimprovements over traditional reuse-only CBP systems from its use of\ntechniques that exploit a high-performance parallel memory of cases.\nCAPER takes a memory-intensive approach by making frequent use of\nmemory during all phases of planning and by using large casebases,\nwhich can be automatically seeded. Because the parallel retrieval\nmechanisms scale to real-world sized casebases of thousands of plans,\nmemory does not have to be pre-indexed and thus retrieval is more\nflexible. Detailed queries can be used to match cases, which are\nstored using an expressive, graph-structured case representation\nscheme.\n\nPlan adaptation in CAPER borrows techniques from generative planning,\nsuch as the use of plan validations, which capture dependencies in a\nplan, and plan composition. These techniques are incorporated into a\nreuse-only CBP framework for a more principled approach to adaptation\nthan in many reuse-only CBP systems. CAPER can also use its flexible\nretrieval mechanisms and case representations to retrieve patch or\nsubstitute plans from memory.\n(Also cross-referenced as UMIACS-TR-95-112)