Computing Reviews

Robust planning with incomplete domain models
Nguyen T., Sreedharan S., Kambhampati S. Artificial Intelligence245 134-161,2017.Type:Article
Date Reviewed: 09/15/17

To develop a plan, a model is constructed. In the model, it is necessary to have complete knowledge about the states, functionalities, and dynamics of the agents in the domain, system, or scenario. These necessities cannot be afforded without complete information about the environment being considered. Sometimes the available information is not complete or is supposed to be acquired gradually when studying the system. Planning based on an incomplete model and learning to improve the completeness of the domain models are the pivotal topics of the paper.

Robust modeling for domains with incomplete information is normally threatened with different risks. Devising methodologies to alleviate this issue is the main stream of investigation in this paper. Conformant probabilistic planning and heuristic search methods are two major approaches to synthesize robust plans.

Operations, predicates, annotations, preconditions, and subjective probability are the mathematical framework concepts used for a decision-theory-based discussion of the proposal; in the incomplete models, annotations for unknown points are utilized to substitute possible preconditions or actions, moving the model to a more favorable state. The technical discussion starts with a pseudolanguage definition that includes the set of items and operations. The authors discuss transition functions for approaching a complete model from incomplete ones. Exponential growth of the possible plans generated, because of the number of possible preconditions and effects, makes the transition function theory an inefficient method. At first refinement, a robustness measure for plans is devised. In the robustness methodology, the probability mass function of successful completion of an action model, based on a sequence of actions toward achieving the goal, is estimated. After counting the salient problems for robust planning, including the compilation approach and heuristic search, issues of assessing plan robustness and synthesizing a robust plan are comprehensively investigated. Finally, a successful plan tracing method to improve plan robustness is provided.

The paper provides a good discussion and rich presentation of the material. It also uses examples to facilitate understanding and schematic graphs to describe the subjects, which are accompanied by useful related scientific explanations.

Reviewer:  Mohammad Sadegh Kayhani Pirdehi Review #: CR145542 (1711-0747)

Reproduction in whole or in part without permission is prohibited.   Copyright 2024 ComputingReviews.com™
Terms of Use
| Privacy Policy