PROCEEDINGS OF THE FOURTH EUROPEAN WORKING SESSION ON LEARNING

• Main Authors: Morik, Katharina (Author), Sallantin, Jean (Author), Quinqueton, Joël (Author)
• Format: Book
• Language: English
• Published: London Pitman 1990; ©1988
• Subjects: Learning, Congresses; Learning, Psychology of Congresses
• Online Access: Click Here to View Status and Holdings.

 This paper represents a synthesis of ideas from qualitative reasoning and explanation-based learn ng Taken together they form a novel approach to planning that relies on plausible inferencing and applies to continuously varying rather than discrete world states. Interestingly, the frame problem skirted and the approach admits some forms of planning under uncertainty. Planning in a domain is very efficient, although learning about the domain can be time consuming. The approach possess a kind of natural reactivity. This paper investigates the application of Explanation-Based Learning (EBL) to planning tacks in continuous demains. To the extent that EBL has been applied to planning, it has been done so almost exclusively using the formalism of situs tion calculus [Chien87, Mitchell85, Segre87, Sim mona89] This is not surprising as situation calculus is still the most common formalism for planning research. Situations are quiescent, permitting no change in truth-value of any formula describing the situation. Actions (or operator applications) are modeled an instantaneous mappings between situs tions. Situation calculus offers a mechanism by which standard inferencing and theorem proving techniques can be applied to problem-solving and planning. Further, it supports the conventional EBL notion of an explanation as a truth-entallment proof, allowing conventional EBL to be performed in planning domains.