|Title||:||Learning and Strengthening Informativeness of Heuristics in Plan Space Planning|
|Speaker||:||Shashank Shekhar (IITM)|
|Details||:||Tue, 5 Jan, 2016 11:00 AM @ BSB 361|
|Abstract:||:||The design of domain independent heuristic functions often throws up experimental evidence that different heuristic functions perform well in different domains. This is often due to the nature of the planning problem, characterized by the degree and nature of interaction between the subgoals. In recent years, learning heuristics has been a matter of investigation. The planning community has observed that techniques for learning heuristic functions have yielded improvements in performance. One approach is to use offline learning to learn predictive models from existing heuristics in a domain dependent manner. These learned models are deployed as new heuristic functions. The learned models can in turn be tuned online using a domain independent error correction approach to further enhance their informativeness. The online tuning approach is domain independent but instance specific, and contributes to improved performance for individual instances as planning proceeds. Consequently it is more effective in larger problems.
We focus on plan space planning and endeavor to enhance the performance of a Partial Order Causal Link (POCL) planner by learning from old heuristics in a supervised manner. We then perform online tuning to minimize the error associated with the offline learned models thus enhancing their informativeness. Our evaluation shows that the learning approaches scale up the performance of the planner over standard benchmarks, specially for larger problems.