Title of Presentation: Human-Robot Teams for Large-Scale Assembly

Primary (Corresponding) Author: Reid Simmons

Organization of Primary Author: Carnegie Mellon University

Co-Authors: Sanjiv Singh, Fred Heger, Laura Hiatt, Seth Koterba, Nik Melchior, Brennan Sellner


Abstract: Autonomous assembly of large structures in orbit and on planetary surfaces will require teams of robots.  Much like any capable human work crew, heterogeneous capabilities will be necessary.  That is, the teams must consist of a variety of agents, some specialized for certain tasks, and others more generalized. We have been developing the architectural framework and software tools necessary to coordinate multiple robots performing complex tasks such as the assembly of multi-element structures.  Since the number of possible failure modes is high when many robots work on long sequences of actions, our architecture allows human operators to cooperatively manage tasks.  We have proposed that complex tasks such as these require a paradigm, which we term "sliding autonomy," in which an operator does not need to control the robots directly but can step in when a problem is detected, or when the robots themselves determine that human assistance is needed.  In previous work over the past 5 years we have developed several examples of structure assembly using a team of robots and an operator.  We have shown that our architecture, which allows autonomy to "slide" between the robots and the operators, helps obtain a higher measure of both efficiency and reliability than is possible when the  robots are either operating autonomously or when they are under the direct (teleoperated) control of an operator.

In current work, we are in the process of demonstrating these ideas in collaboration with the Space Systems Laboratory at the University of Maryland.  Our system will enable the autonomous assembly of the space truss structure, using the multi-arm Ranger robot.  The structure is a tetrahedron, consisting of 4 nodes and 6 beams, that was assembled several times in the Space Shuttle's bay while in orbit as a test of astronauts' abilities to assemble and maintain structures.  We expect to demonstrate a system that executes and monitors complex, multi-arm assembly operations, freeing a remote operator from managing low-level control during normal operation, while allowing the operator to step in when problems arise.  Our aim is to quantify the benefits of this hybrid mode of control in terms of reliability (failure rates) and efficiency (time to completion).