Of Plans, Programs, and Automata: 101 things you can do with AI Automated Planning
The field of Artificial Intelligence Automated Planning has seen significant advances in the last 15 years, largely as a result of innovations in planning-specific heuristic search and SAT techniques.
Much of the focus has been on so-called classical planning systems which assume complete information about the initial state of a system at the outset of planning, deterministic actions, and a final state goal.
Unfortunately, many interesting real-world problems violate classical planning assumptions. In this talk, I will discuss planning with temporally extended goals, constraints, and preferences, specified in Linear Temporal Logic (LTL) and via programs specified in an Algol-like language called Golog, which together capture the expressivity of finite state automata. A main focus of this talk will be on how to leverage these expressive languages to help guide heuristic search. We have explored these techniques in a diversity of problem settings from web service composition to verification and concurrent test generation. This talk should be of some interest to anyone who is concerned with reachability in dynamical systems. This is joint work with Jorge Baier, Christian Fritz, and Shirin Sohrabi.