Title: A Robot Manipulation Framework of Novel Objects for Elementary Household Tasks

Ana C. Huamán Quispe

Robotics PhD student

School of Interactive Computing,

College of Computing

Georgia Institute of Technology

www.cc.gatech.edu/~ahuaman3

Date: Tuesday, October 13th, 2015

Afternoon session time: 4:15 pm - 6:00 pm (EST)

Location: TSRB 226

Committee:

Prof. Henrik I. Christensen (Advisor), School of Interactive Computing, Georgia Institute of Technology

Prof. C. Karen Liu, School of Interactive Computing, Georgia Institute of Technology

Prof. Andrea L. Thomaz, School of Interactive Computing, Georgia Institute of Technology

Prof. Charles Kemp, School of Biomedical Engineering, Georgia Institute of Technology

Prof. Peter K. Allen, Department of Computer Science, Columbia University

----
Prof. Mike Stilman, School of Interactive Computing, Georgia Institute of Technology

Abstract:

Through the years, robot manipulation research has widened its application field  from structured industrial environments to household scenarios. For a robot manipulator  to operate  in the latter, it must be able to handle diverse type of objects in order to perform equally diverse manipulation tasks. Although theoretically, a robot could take full advantage of online databases to identify objects and access their corresponding grasping information, we claim that a robot should also have the capability to perceive a novel object, create a suitable representation of it, and generate a manipulation plan in an efficient and online manner. In this proposal, we present a framework to manipulate novel objects to accomplish elementary household tasks. We present our work on object representation based on superquadrics and its direct relation to fast grasp generation. Second, we present an algorithm to generate manipulation plans for pick-and-place tasks by selecting a grasp that maximizes a metric based on a human heuristic (end-comfort effect). Third, in order to define grounded elementary household tasks, we present a study based on human dexterity tests to define benchmark tasks to evaluate our system. Finally, we explore handover tasks, as a final application to our approach, necessary to operate in household environments. We validate our results with experiments on 2 robotic platforms.