My thesis work deals with planning in uncertain domains with durative actions and resources. Incorporating such uncertainty into the plan itself is very difficult to do, and so a common way of approaching such problems is to plan deterministically and simply replan as needed during execution. Doing so, however, has obvious drawbacks that make the approach suffer with respect to optimality.

In order to help improve the performance of such deterministic systems, I am working to develop a framework that incorporates domain probabilities back into the planning cycle, by considering them at execution time as plans are being manage to adapt to the realities of execution. One example of how to do so is to base the decision of whether to replan during execution time based on the probability that something is actually likely to go wrong, not whether the deterministic plan thinks there will be a problem. I call this approach Probabilistic Plan Management. Probabilistic plan management has already been shown to lessen the time spent replanning during execution, and improve the quality of executed plans, by preventing wasted planning effort and better focusing planning efforts in such ways.