Title: Robogyver: Autonomous Tool Macgyvering for Inventive Problem Solving
Lakshmi Velayudhan Nair
Robotics Ph.D. candidate
School of Electrical and Computer Engineering
Georgia Institute of Technology
Date: Friday, September 25th, 2020
Time: 11 AM to 1 PM (ET)
Remote only because of GT COVID-19 guidelines.
Dr. Sonia Chernova (Advisor) – School of Interactive Computing, Georgia Institute of Technology
Dr. Brian Magerko – School of Literature, Media, and Communication, Georgia Institute of Technology
Dr. Devi Parikh – School of Interactive Computing, Georgia Institute of Technology
Dr. Mark Riedl – School of Interactive Computing, Georgia Institute of Technology
Dr. Christopher Atkeson – Robotics Institute, Carnegie Mellon University
Robots that are situated in the real world are often faced with unforeseen situations that require them to adapt and improvise to be more useful. Particularly in the context of using tools, there may be situations where a robot does not have access to the tools it needs for completing a task. While humans show remarkable improvisation capabilities, similar skills are beyond the scope of robots today. In order to address these scenarios, a resourceful robot should be able to use whatever objects are available to it, in order to replace the missing tool. We refer to this process as “tool macgyvering”. Tool macgyvering can be achieved by either substituting the missing tool with an object (tool substitution) or constructing a replacement tool by combining multiple objects (tool construction). This thesis focuses on the problem of tool macgyvering, and contributes: (1) a formalization of three levels of tool macgyvering that highlights the levels of complexity involved in tool macgyvering problems; (2) novel algorithms for tool construction through shape, material and attachment reasoning, where attachment refers to the different ways in which objects can be combined; (3) a novel algorithm for tool substitution using shape and material reasoning; (4) a novel framework that performs tool macgyvering through arbitration of substitution and construction, to enable a robot to effectively decide between the two solutions; and (5) a novel algorithm to perform tool macgyvering in task planning, to enable a robot to leverage existing planning algorithms to perform tool macgyvering efficiently.