1 / 1

Director: Prof. Maja J Matarić Associate Director: Prof. Gaurav S. Sukhatme

E. A. C. B. A. t1 E. t2 E. t1 A. t1 B. t2 B. t1 C. t1 D. t2 D. t1 C. t2 A. Effects Beh 1…k {1/0}. Test task specific preconditions. Effects Beh i {1/0}. Task specific preconditions. if met. if met. Test world preconditions. Test world preconditions. Abstract behavior.

tawana
Download Presentation

Director: Prof. Maja J Matarić Associate Director: Prof. Gaurav S. Sukhatme

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. E A C B A t1E t2E t1A t1B t2B t1C t1D t2D t1C t2A Effects Beh1…k {1/0} Test task specific preconditions Effects Behi {1/0} Task specific preconditions if met if met Test world preconditions Test world preconditions Abstract behavior if met Perform actions A C A E Primitive behavior Perform actions Standard behavior structure Abstract/primitive behavior structure Network links types (sequential preconditions): Postconditions true Behavior A Behavior A Behavior A Behavior active Behavior B Behavior B Behavior B t1A t1B t2B t2A t1A t1B t2A t2B t1A t2A t1B t2B Permanent Enabling Ordering Primitive Behavior Abstract Behavior Network Abstract Behavior “Expanded” representation of a NAB Network link (ordering, enabling, permanent) Activation link Observation Relation Link J K J Before K Ordering J K J Meets K Enabling J K J Overlaps K J J Includes K Permanent K J J Ends K K J J Starts K No relation K J K J Equals K AN ACTION-BASED FRAMEWORK FOR LEARNING FROM DEMONSTRATION IN HUMAN-ROBOT DOMAINS Director: Prof. Maja J Matarić Associate Director: Prof. Gaurav S. Sukhatme Founder: Prof. George A. Bekey Monica N. Nicolescu and Maja J. Matarić (monica|mataric@cs.usc.edu) http:||robotics.usc.edu|~monica GOALS APPROACH • Automate the process of robot controller design : • Complexity of the robot’s tasks (sequencing) • Robustness and real-time response properties • Modularity of the underlying architecture • Reusability of controller components • Support for complex task learning • Teaching by experienced demonstration • The robot performs the task during demonstration and perceives the task through its own sensors • Mapping observations to the robot’s own set of actions • A Hierarchical Abstract Behavior-Based Architecture • Representation & execution of complex, sequential, hierarchically structured tasks • Sequential & opportunistic execution THE ARCHITECTURE • Separate sensing (precondition checking) from actions into abstract/primitive behaviors. • allows for a more general set of activation conditions • Embed abstract representation of the behavior’s goals • the task specific preconditions are tested via behavior links • Tasks are represented as (hierarchical) behavior networks. Generic network • Execution: activation spreading + precondition checking • Activation level (of a behavior): The number of successor behaviors in the network that require the achievement of its postconditions • Behavior selection: a behavior is active iff : ( It is not inhibited ) and • ( Its controlled actuators are available) and • ( Activation level 0 ) and • ( Allordering constraints = TRUE ) and • ( All permanent preconditions = TRUE ) and • (( All enabling preconditions = TRUE ) or • ( the behavior was active in the previous step )) THE LEARNING PROCESS • Teacher-following strategy • The robot has a set of basic skills • Teacher signals moments in time relevant to the task • Mapping observations to the known effects of the robot’s own actions Teacher signals EXPERIMENTALVALIDATION ALGORITHM: • The intervals Ik occurred during the demonstration • Behavior set: PickUp & Drop colored objects, Track colored targets • Learning in clean/cluttered environments, from human/robot teachers • A task with long sequences • A slalom task • An object transport task • A “gate-traversing” task An object transport task: Teacher demonstration: • Create network back-bone A C B E A • Generate behavior links Learned network: • Example network REFERENCES [1] Monica N. Nicolescu, Maja J Matarić, "A hierarchical architecture for behavior-based robots", First International Joint Conference on Autonomous Agents and Multi-Agent Systems, July 15-19, 2002 [2] Monica N. Nicolescu, Maja J Matarić, "Learning and Interacting in Human-Robot Domains",Special Issue of IEEE Transactions on Systems, Man, and Cybernetics, Vol. 31, No.5, Pages 419-430, September, 2001. [3] Monica N. Nicolescu, Maja J Matarić, "Experience-based representation construction: learning from human and robot teachers", IEEE/RSJ International Conference on Intelligent Robots and Systems, Pages 740-745, Oct. 29 – Nov 3, 2001

More Related