Kaijen Hsiao

I'm a grad student in the Computer Science and Artificial Intelligence Lab at the Massachusetts Institute of Technology, in the Learning and Intelligent Systems Group run by Professors Leslie Kaelbling and Tomas Lozano-Perez.

At the moment I'm working on robust manipulation under uncertainty, which means that I want robots to be able to grasp objects and put them down accurately even when they have noisy estimates of where the objects are, or what their hands are feeling, or even the precise details of the shapes of the objects. Normal, precise grasp planning methods often result in failing to grasp or even knocking objects over when the world is noisy. This is something that people often solve using compliant manipulators that just glom over the object, but for some tasks, precise grasps are required (such as grasping handles, using tools, or doing assembly operations). In our case, we model the world as a POMDP (Partially Observable Markov Decision Process), and try various complexities of planning such as solving the full POMDP (when tractable) or, more recently, choosing appropriate fixed, pre-planned trajectories based on the current belief state, which we then try to execute robustly (what doesn't kill you, gives you information. Try try again).

Here are links to my publications:
"Reactive Grasping Using Optical Proximity Sensors," Kaijen Hsiao, Paul Nangeroni, Manfred Huber, Ashutosh Saxena, and Andrew Ng, ICRA 2009
Full-system videos (Barrett Arm/Hand with optical sensors) here,
The controller with just normalized raw IR values (works just as well): Successes, Failures due to surface properties, Failures due to not having normals

"Robust Belief-Based Execution of Manipulation Programs," Kaijen Hsiao, Tomas Lozano-Perez, and Leslie Pack Kaelbling, WAFR 2008
Talk slides here; Videos (simulated Barrett Arm): Grasping the center of a box, Grasping a cup from above, Grasping a Brita pitcher

"Grasping POMDPs: Theory and Experiments," Ross Glashan and Kaijen Hsiao and Leslie Pack Kaelbling and Tomas Lozano-Perez, RSS Manipulation Workshop: Sensing and Adapting to the Real World, 2007.
Videos (real Barrett Arm): One finger finding a stepped block, Two fingers finding a box

"Grasping POMDPs," Kaijen Hsiao and Leslie Pack Kaelbling and Tomas Lozano-Perez. ICRA, 2007.
Videos (simulated robot): Grasping blocks, Grasping stepped blocks

"Imitation Learning of Whole-Body Grasps," Kaijen Hsiao and Tomas Lozano-Perez. IROS, 2006.
Video (simulated humanoid robot): Grasping random objects

"Evolving Simulated Mutually Perceptive Robots for Combat," Michael O'Kelly and Kaijen Hsiao. ALIFE 2004.
Videos of the evolved robots are at the BubbleGene webpage

"Passive Acoustic Knock Tracking for Interactive Windows," Joseph A. Paradiso, Che King Leo, Nisha Checka, Kaijen Hsiao. ACM CHI2002, Minneapolis, MN.

"Passive Acoustic Sensing for Tracking Knocks Atop Large Interactive Displays," Joseph A. Paradiso, Che King Leo, Nisha Checka, Kaijen Hsiao. IEEE International Conference on Sensors, 2002.

Here are some abstracts of past work:
Grasping POMDPs
Imitation Learning of Whole-Body Grasps

Here is my resume.

And here is a page with pictures of some of my crafts.