Hierarchical Task and Motion Planning in the Now. Kaelbling, Lozano-Perez. ICRA 2011.


  1. Considers task and motion planning
  2. “It is aggressively hierarchical; it makes choices and commits to them in a top-down fashion in an attempt to limit the length of plans that need to be constructed, and thereby exponentially decrease the amount of search required.”
  3. Uses continuous geometric representations, does not require a-priori discretizations
  4. Consider realistic planning problems that aren’t pathological – where mistakes aren’t catastrophic, there is enough freedom of movement that planning isn’t terribly hard, and that optimality is not necesary.
  5. “Current symbolic task planners and geometric motion planners have complimentary strengths… a task planner could decide that the living room needs to be traversed, regardless of the detailed arrangement of its furniture. Motion planners deal beautifully with geometry, but not with non-physical aspects of the domain; they can plan how to get to the phone but not decide that a phone call needs to be made.”
  6. The idea is that due to stochasticity, and the fundamental inability to create accurate plans deep into the future due to computational restrictions, planning is done in a particular way
    1. Figure out a set of abstract steps that have to be accomplished
    2. Plan out in detail only the current step in the plan, be robust to failures
  7. There is a possibility that the plan may not be feasible due to the fact that it is only done approximately (especially in terms of items aside from the initial step).
    1. They attempt to constrain the plans to be serializable such that any solution for the first step will allow for some solution to subequent steps
  8. “If, for some reason serializability fails, the  we formulate an interleaved plan for achieving the effects of both steps; as long as actions in the environment are ultimately reversible, then any goal can be achieved, at the expense of sub-optimality in the behavior.”
  9. This approach is similar to: manipulation planning, integration of symbolic and motion planning, and hierarchical planning (with relevant citations).
  10. Planning is done at a symbolic level, and then if it cannot be executed directly then a low-level plan is constructed, then a very coarse, conservative motion planning is done, and once that succeeds it is planned again at a finer grain
  11. Fluents are: In, Overlaps, Clear (is clear?), Holding, Clean
  12. Dont completely represent the world in terms of fluents because that may be too big, but has a representation that can give answers to those queries as needed (for example, a geometric representation is more compact but can produce answers the queries)
  13. Much of the approach does not commit to a particular representation but they describe their choices for their example – some things have a couple of levels of symbolic representation followed by a low-level motion planner
  14. <Not really getting anything out of this based on my current objectives so stopping here.>
Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: