IDA Assistance for Mixed-Initiative Planning

In mixed-initiative planning, a human planner collaborates with an automated system to build a plan of action. In existing approaches, the participants incrementally add constraints or modi cations to a single plan structure until an acceptable solution is found. We propose instead to treat the process as a form of data exploration, in which sets of plans and their relationships are the objects of interest, rather than an individual abstract plan. The system assists the human planner in this process by autonomously constructing simple statistical descriptions of patterns in sets of plans. Its analysis is supported by heuristic knowledge about the physics of a planning environment, knowledge that contributes to a limited form of strategic statistical reasoning.