Plan Recognition in HTN Planning

In plan and goal recognition, a partially executed plan is observed (e.g., the beginning of some plan), and the task is to figure out to which complete plan it belongs, i.e., which steps are still missing (called plan recognition). This goes hand in hand with identifying what the overall task is that the agent plans for (called goal recognition). In HTN planning, the set of possible plans is given in terms of an HTN model, often called a plan library.

  • Plan Recognition via Encoding into HTN Planning:
    • URL: will be provided soon.
    • Further info: This approach encodes the plan and goal recognition problem as an HTN problem so that the solution to the new HTN problem contains the executed plan prefix plus the still missing plan. It also encodes which plan (initial compount task) the agent pursued. Approach was described at ICTAI 2018 (won the best paper award). Details will be provided soon.