Generalized Task Structure Learning for Collaborative Multi-Robot/Human-Robot Task Allocation

In order to allow the learned task to be generalized, the structure of the task must be flexible to allow for these variances in grasping capabilities for heterogeneous teams of robots. Therefore, it is not sufficient to give a single, rigid task structure to a team of heterogeneous robots and assume that they will be able to complete the task without taking into account their own skill-sets. Instead of manually crafting a different task structure for each robot in the team, we propose a method in which we are able to learn the underlying task structure from a set of human demonstrations which can be transferred to a robot and modified online with its own suitability for each part of the task through the utilization of our previous developed hierarchical control architecture.

By learning the structure of the task instead of the exact details or motor movements of the task, we are able to impart the individual skills of each robot within the task structure to ensure that all robots are performing the same task but with respect to their own suitability. The contributions of the proposed work are: (1) an extension of our previously developed control architecture to allow for conditional constraints, (2) a novel method which is able to take sequences of demonstrations and learn a hierarchical task representation, and (3) an end-to-end system which is able to take demonstrations provided by a human and transfer the learned skills to a team of heterogeneous robots.

Role: Lead graduate researcher/ main mentor of undergraduate students.

Associated Publications:

• Ph.D. Dissertation:
  • Janelle Blankenburg, “Generalized Task Structure Learning for Collaborative Multi-Robot Task Allocation”, University of Nevada, Reno, NV, May 2020 (expected)
• In Preparation Conference Proceedings:
  • Janelle Blankenburg, Andrew Palmer, Monica Nicolescu, David Feil-Seifer. “Generalized Hierarchical Task Learning Through Human Demonstration.” In IEEE-RAS In International Conference on Intelligent Robots and Systems (IROS), Oct 2021 (in prep).
  • Eloisa Burton*, Janelle Blankenburg, Monica Nicolescu, David Feil-Seifer. “Interdependence Constraint for Collaborative Multi-Robot Task Allocation Using a Distributed Control Architecture.” In IEEE-RAS International Conference on Humanoid Robots (Humanoids), Dec 2021 (in prep).