1 / 11

Integrating a Closed World Planner with an Open World Robot: A Case Study

Integrating a Closed World Planner with an Open World Robot: A Case Study. Kartik Talamadupula J. Benton Subbarao Kambhampati Dept. of Computer Science Arizona State University. Paul Schermerhorn Matthias Scheutz Cognitive Science Program Indiana University. Urban Search & Rescue.

ivy
Download Presentation

Integrating a Closed World Planner with an Open World Robot: A Case Study

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Integrating a Closed World Planner with an Open World Robot: A Case Study Kartik Talamadupula J. Benton Subbarao Kambhampati Dept. of Computer ScienceArizona State University Paul Schermerhorn Matthias Scheutz Cognitive Science ProgramIndiana University

  2. Urban Search & Rescue Wounded people in rooms Soft Goal Report the locations of wounded people Human-Robot team Robot starts at beginning of hallway In communication with the human Hard Goal Reach the end of the hallway

  3. How do you make a deterministic closed-world planner believe in opportunities sans guarantees? Open World Quantified Goals Partial Satisfaction Planning (PSP) Sensing and Replanning Planner Robot Closed World Open World Limited SensingPlanner guides robot in a limited way Over SensingRobot senses its way through the world Under SensingClosed World Model

  4. Bias the Planner’s Model • Endow the planner with an optimistic view • Assume existence of objects and facts that may lead to rewarding goals • e.g. the presence of a victim in a room • Create runtime objects • Add to the planner’s database of ground objects • Plans are generated over this reconfigured potential search space planners Some people see things as they are and say why? I dream things that never were and say why not? --(mis)attributed to Robert Kennedy (who quoted Bernard Shaw) Our planner dreams s

  5. Open World Quantified Goals (OWQGs) • Goals that allow for the specification of additional information • To take advantage of opportunities (:open  (forall?r – room (sense ?p – person (looked_for ?p ?r) (and (has_property ?p wounded) (in ?p ?r)) (:goal (and (reported ?p wounded ?r) [100] - soft)))) Quantified Object(s) Sensed Object Closure Condition Quantified Facts Quantified Goal

  6. Partial Satisfaction Opportunities retain their bonus nature SOFT GOALS Not enough time (or other metric resources) to fulfill all goals QUANTIFIED GOALS Sensing has a cost NET BENEFIT Rewards from object existence

  7. Planning to Sense Sensing to Plan • Sensing is expensive … • Cannot be done at every step • Planner needs to direct the architecture on: • when to sense • what to sense for • Planning to sense in a goal-directed manner • Output all actions up to (and including) any action that results in sensing closure • Tempers the optimism created in the planner by the OWQGs Updated State Information Plan Goal Manager Monitor Planner Plan Problem Updates Robot

  8. Evaluation • Evaluated on the USAR scenario • Experimental Setup • Robot in a corridor; hard goal to reach the end. • Three rooms with green, blue and no box respectively. • Hard goal has a timed deadline • Planner reports wounded people in rooms

  9. Limitations & Extensions LIMITATIONS eXTENSIONS • Lack of support for uncertainty • Uncertainty about fact distribution • E.g. Wounded people more likely in bathrooms • Determinization is too optimistic • Object existence assumed • Wumpus World: A wumpus in every room? • Augment OWQGs to handle probabilities • Generate more than one outcome (FF-HOP) • Precautionary Planning • Replan for execution failures • Add precautionary measures to deal with unrecoverable failures

  10. Related Work • Design rewards such that reactive behavior leads to the optimal goal [Marthi 2007] • Sprinkle donuts on the road liberally to lure Homer Simpson into the library • Local Closed World statements [Etzioni et al. 1997] • Completely open world • Parts of the planner’s representation of the world closed via LCW statements • Our approach is complementary • OWQGs open parts of a closed world

  11. Summary How did we make a deterministic closed-world planner believe in opportunities (without guarantees)? Open World Quantified Goals Planner Robot Closed World Open World Sensing & Replanning 1. Temper planner’s optimism 2. Replan for unexpected states and new information Partial Satisfaction Planning 1. Maintain bonus nature 2. Quantified goals 3. Limit sensing actions

More Related