PR2 Beta Sites: Spotlight on UC Berkeley

LaundryPR2 Beta Program: A Platform for Personal Robotics

The University of California, Berkeley already has a history with the PR2. Last winter, Jeremy Maitin-Shepard spent his nights working on a PR2 at Willow Garage. At the end of this project, Jeremy had the PR2 folding towels with a success rate of 100%. Now, UC Berkeley is getting its own PR2, and we're excited to see what else the team will accomplish. Berkeley's ambitious plans for the PR2 include manipulation of deformable objects, hierarchical planning, perception, and learning from demonstrations.

Towels were only the beginning of Berkeley's work with non-rigid materials and laundry. Berkeley's researchers will use the next two years to work on taking it to a whole new level: doing the full laundry cycle, from dirty clothes in a laundry basket to clean, folded clothes. This will present multiple, more difficult challenges. One of the challenges in folding towels was identifying the correct places to grab a crumpled towel and straighten it into a flat, foldable towel. With clothes, Berkeley will have to improve their techniques to handle more difficult and varied shapes. The PR2 will also have to operate with greater robustness and complete new tasks in order to finish the full laundry cycle.

Last summer, UC Berkeley's Jason Wolfe studied hierarchical planning with the PR2 during his internship at Willow Garage. Humans use this type of planning everyday to make the most of their movements. If you have to carry the laundry to your room and replace the batteries in your alarm clock, you'll probably plan to swing by the kitchen drawer for batteries before picking up the laundry basket and heading to the bedroom. You could make two separate trips, but your time is more efficiently used if you're able to plan your actions more optimally. The same kind of task planning will make robots more efficient, and UC Berkeley will work on this challenging research area with their new PR2.

In the area of perception, the Berkeley team will work on improving the PR2's ability to find and interact with real-world objects. Some of the research areas Berkeley will work on include recognizing transparent objects, like glasses, finding people, and determining the correct way to grasp objects.

Lastly, Berkeley plans to use the PR2 for learning from demonstrations. Programming robots can be time-consuming and requires expert programmers. What if you could teach a robot simply by demonstrating what you want it to do? Berkeley will work on this challenge and teach the PR2 tasks, like basic assembly of objects.

Berkeley teamThe Team

The UC Berkeley team brings together expertise over a wide range of relevant fields.

The team includes an excellent group of graduate students and postdocs, including the following: Mario Fritz, Haomiao Huang, Warren Hoburg, Judy Hoffman, Sergey Karayev, Jeremy Maitin-Shepard, Stephen Miller, Mathieu Salzmann, Pranav Shah, Arjun Singh, Hyun-Oh Song, Jie Tang, Ramanrayan Vasudevan, Michael Vitus, and Jason Wolfe.


Below is a video of Arjun Singh presenting the UC Berkeley proposal to the rest of the PR2 Beta Program participants. You can download the slides as PDF.