The CL-Buridan System: Probabilistic Planning with Incomplete Information

Planning with Incomplete Information for Robot Problems

Volume

Planning with Incomplete Information for Robot Problems

Contents

This paper describes CL-BUItIDAN an implemented planner for problem domMns in which the agent is uncertain about the initial world state and the effects of its own actions, but has sensors that allow it to improve its state of information. The system uses a probabilistic semantics to represent incomplete information, and provides for actions with informational as well as causal effects. The action representation allows an action’s causal and informational effects to be freely mixed, and can represent sensors whose informational content is noisy and state dependent. Information obtained at run time can be exploited by conditional and looping constructs in the plan language. This paper describes the basic structure of the C L-B V RIDAN representation and algorithm, and also provide an analysis of assumptions about target problem domains that would make them an appropriate (or inappropriate) choice for using this technology build a problem-solving agent.