The most common representation formalisms used by AI planning algorithms are descriptive models. They abstractly describe what the actions do and are tailored for efficiently computing the next state(s) in a state transition system. However, the actor uses operational models that describe how to do things, with rich control structures for closed-loop online decision-making. There have been several approaches that combine planning with descriptive representations and acting with operational models. However, they have several drawbacks, in particular, with respect to the development and consistency verification of models.
We propose to develop an integrated online acting-and-planning algorithm in which both planning and acting use the same operational models. We also aim to prove formal properties of the algorithm such as correctness and time complexity. The operational models are written in a general-purpose hierarchical task-oriented language offering rich control structures. The acting component is inspired by the well-known PRS system, except that instead of being purely reactive, it can get advice from the planner. We have already developed two planning algorithms, APEplan and RAEplan and their formal properties. They plan by doing Monte Carlo rollout simulations of the actor's operational models. Our current experiments show significant benefits in the efficiency of the acting and planning algorithms. Our plan for future work is to make our planner more efficient and accurate and to deploy our acting-and-planning algorithm in a real-world domain.
Dept. rep: Dr. Tom Goldstein
Members: Dr. Jordan Boyd-Graber