Adaptive Sampling Applied to Planning with Hindsight Optimization

Generating plans in stochastic planning enviironments is a difficult and time consuming process even with state of the art planners such as FF-Hindsight. This research applies a series of adaptive sampling methods around FF-Hindsight to improve the quality of plans returned by the planner as well as allowing the planner to modify the time it takes to make a decision on an action and base it on the diffculty of the problem. By creating a pool of samples and distrbuting these based on a variety of selection methods, the planner can be more capable of solving a variety of planning problems.