Learning by Doing: Realized Low-level Manipulation Skills as the Foundation for High-level Human-robot Collaboration

Abstract

The concrete goal of this program is to achieve robots that are cognitively compatible with human team members, in their ability to execute the skills that will be expected of them. In turn, this will enable an approach more closely resembling goal-based commanding in human-robot teams. Still, we believe that many of the steps presented here will advance the ?eld towards increasing levels of autonomy.A common challenge to training robot team members is the need to collect large amounts of demonstrations, exclusively for the bene?t of the robot, and for each task that is expected. Such an approach simply does not scale. We present here a method for the robot to learn (via teleoperation) skills that are more general than tasks, and which can then be executed autonomously. Tasks are then executed under the supervision (and high-level commands) of human team members. Additional data, for both low-level skill execution and high-level skillsequencing into tasks, can be collected during this process, building towards the ultimate goal of obtaining full, task-level autonomy. The program proposed here starts from the ground up, with realization of motor skills; as such, going all the way to full task autonomy is outside of the proposed scope. However, these are important steps on this path.

Document Details

Document Type
DoD Grant Award
Publication Date
Jan 23, 2019
Source ID
N000141912062

Entities

People

  • Matei Ciocarlie

Organizations

  • Office of Naval Research
  • Trustees of Columbia University in the City of New York
  • United States Navy

Tags

Fields of Study

  • Computer science

Readers

  • Instructional Design and Training Evaluation.
  • Robotics and Automation.
  • Team-Based Human-Centered Cognitive Task Decision Making and Information Performance.

Technology Areas

  • AI & ML
  • AI & ML - Autonomous Systems
  • Autonomy