The goal of this project is to develop an action-based framework for interaction between humans and robots that facilitates the transfer of knowledge (from humans to robots and between the robots themselves) and communication strategies that extend the ones typically used in the mobile robotic systems domain. We consider that the robots are equiped with a set of basic skills (in the form of behaviors) which constitute the basis for future task learning and communication.

Our methods for human-robot interaction are focused on learning hierarchical representations of complex tasks from a small number of robot experienced demonstrations, with minimal or no feedback from a human demonstrator. The approach relies on a hierarchical abstract behavior architecture, an extension of the robust and flexible behavior-based systems (BBS), which allows for the representation, learning and execution of complex, hierarchically structured tasks. We have developed an on-line algorithm that constructs hierarchical behavior network representations of robot tasks based on a robot's experienced demonstrations (videos) with a human or another robot teacher.

Our methods for robot-human interaction are focused on developing an action-based communication framework (videos) in which a robot can convey its intentions by suggesting them through actions rather then communicating them through conventional signs, sounds, gestures, or marks with previously agreed-upon meanings. We consider the mobile robot's actions a vocabulary that it could use to induce a human to assist it for parts of tasks that it is not able to perform on its own.

Experimental setup	Description	Videos
	Learning to visit a number of targets in a certain order	Human demonstration [.AVI] The robot reproducing the learned task [.AVI]
	Learning to traverse "gates" and move objects from a source place to a destination	Human demonstration [.AVI] The robot reproducing the learned task [.AVI] Human demonstration [.AVI] The robot reproducing the learned task [.AVI]
	Learning to slalom	Human demonstration [.AVI] The robot reproducing the learned task [.AVI]

Experimental setup	Description & Videos
	Traversing blocked gates with a person's help [.AVI] Moving inaccessible located objects with a person's help Interacting with an uninterested person[.AVI] Interacting with an unhelpful person[.AVI] Interacting with an helpful person[.AVI]

• Monica Nicolescu, Maja J Mataric, "Learning and Interacting in Human-Robot Domains", to appear in Special Issue of IEEE Transactions on Systems, Man, and Cybernetics, Part A: Systems and Humans , 2001.[PS version], [PDF version]

• Monica Nicolescu, Maja J Mataric, "Experience-based representation construction: learning from human and robot teachers", Proceedings, IEEE/RSJ International Conference on Intelligent Robots and Systems, Maui, Hawaii, USA, October 29 - November 3, 2001. [PS version]

• Monica Nicolescu, Maja J Mataric, "Experience-based learning of task representations from human-robot interaction", Proceedings, IEEE International Symposium on Computational Intelligence in Robotics and Automation, Banf, Alberta, CANADA, July 29 - August 1, 2001.[PS version], [PDF version]

• Monica Nicolescu, Maja J Mataric, "Extending Behavior-Based Systems Capabilities Using An Abstract Behavior Representation", Working Notes of the AAAI Fall Symposium on Parallel Cognition, pages 27-34, North Falmouth, MA, Nov 3-5, 2000. [PS version], [PDF version]

• Monica Nicolescu, Maja J Mataric, "Learning Cooperation From Human-Robot Interaction", Proceedings, 5th International Symposium on Distributed Autonomous Robotic Systems (DARS), pages 477-478, Knoxville, TN, Oct 4-6, 2000. [PS version], [PDF version]

• Monica Nicolescu, Maja J Mataric, "Deriving and Using Abstract Representation in Behavior-Based Systems", Proceedings, Seventeenth National Conference on Artificial Intelligence (AAAI), Student poster, page 1087, Austin, Texas, July 30-August 3, 2000.[PS version], [PDF version]

This work is supported by DARPA Grant DABT63-99-1-0015 under the Mobile Autonomous Robot Software (MARS) program, the National Science Foundation Grant No. 9896322, and by the ONR Defense University Research Instrumentation Program Grant.

Monica Nicolescu
Maja Mataric.