Planning Under Temporal Uncertainty Using Hindsight Optimization

A robot task planner must be able to tolerate uncertainty in the durations of commanded actions and uncertainty in the time of occurrence of exogenous events. Sophisticated temporal reasoning techniques have been proposed to deal with such issues, although few existing planners support them. In this paper, we demonstrate the capabilities of a much simpler technique, hindsight optimization, in which uncertainty is handled by using sampling to generate deterministic planning problems that can be solved quickly. We find that sophisticated temporal reasoning is not required to handle many simple tasks. In comparison with a traditional temporal planner architecture, hindsight optimization is much simpler to implement while staying closely integrated with execution. It serves as a flexible baseline against which more complex methods can be compared.