Matthew T. Mason
Professor, Computer Science and Robotics
I work in robotics. The primary venue for my work is the Manipulation Lab (MLab).
My primary focus is simple hands. I think simple hands are better than complex anthropomorphic hands, both for research purposes and for most applications. The main reason is that the dexterity of a hand is mostly about the brain, not the hand. Our approach mixes a strong dose of machine learning into the traditional brew of AI, control and mechanical design. Probably the first one to read is the 2012 IJRR paper Autonomous Manipulation with a General-Purpose Simple Hand. Also see related papers from ICRA 2013 and RSS 2011, both of which won best student paper awards. The project began with NSF funding and is continuing with funds from NSF and the Army Research Office.
In robotic manipulation lingo, “dexterity” often refers to moving an object in the hand, like when you drop a coin from a fingertip grasp to your palm. However, most research on dexterity assumes a complex anthropomorphic hand. Under the standard assumption, a simple hand cannot shift grasps. But those assumptions are too restrictive. We have demonstrated the mechanical feasibility of dexterous manipulation using a simple hand. Our research is focused on going beyond mechanical feasibility, so that an autonomous robot could employ extrinsic dexterity by planning and controlling the simple hand for a given object and task. The first paper to read is the 2014 ICRA paper (Chavan Dafle et al.), perhaps followed by the Zhou et al. paper which won Best Conference Paper at ICRA 2016, or the upcoming WAFR 2016 paper by Hou et al.
Industrial automation is the ultimate proving ground for robotic manipulation research. This project, sponsored by Foxconn, is focused on automatic assembly of smart phones. The challenges are formidable, since smart phones use very small parts, some floppy and some sticky, and there is unavoidable clutter. We have a forthcoming paper surveying threaded fastening automation, and a recent paper on machine learning applied to classify stages of screw insertion (ISER 2016).