I began working in the Learning and Intelligent Systems Laboratory under Professors Leslie Kaelbling and Tomas Lozano-Perez through the UROP program, which allows undergraduates to work in research labs. I was assigned to work with a graduate student, Jenny Barry, who has been my mentor. I worked with several other undergraduates under Jenny’s supervision on a project using her newly developed for controller. This work was done on Willow Garage’s PR2, a personal robot created as a research platform. With the goal of doing household chores in mind, we began working on the simple task of opening kitchen cabinets. Much of the previous work on this task had focused on a complex planning where a model of the dynamics of the door had to be learned or known. For a kitchen cabinet, the model of the handle’s circular trajectory as the door opened would have to be calculated and executed with minimal error to ensure that the door did not get stuck. Instead, we developed an open-loop hybrid controller. A hybrid controller simply specifies a force along one axis and a desired position along another. This type of controller was useful for opening doors because it allowed us to take advantage of some of the natural physical constraints of the door. Once the robot has successfully grasped the handle, there are only two degrees of freedom. In the initial configuration with the door closed, the desired motion can be approximated linearly outward (in an outward opening door). A position controller can be used in that axis, allowing the opening action to begin. A force controller can then be used in the perpendicular axis. The combination of these controllers approximates the tangential trajectory the handle must take. We tested this approach on several types of cabinets, and in each case this approximation allowed us to successfully open the door without time consuming or computationally intensive planning.
This work proved relevant to another graduate student, Mario Bollini, on his BakeBot project. His project involved the PR2 baking cookies. I used our work on door opening to open the oven for the cookies. This project gave me my first experience with collaboration. I initially worked on the project as a team with other undergraduates; later I was able to extend our work by applying it to an unrelated project. This application exemplified the broader impact of our work. Opening doors is a basic tool in many robotic tasks, and our approach was meant to provide a simple and functional solution. I was fortunate enough present a poster on this work with Jenny at the International Conference on Intelligent Robots and Systems (IROS) in September of 2011. I traveled to San Francisco with Jenny and several other members of our lab.
Planning and Control Under Uncertainty for the PR2
Jennifer Barry, Mario Bollini, Anne Holladay, Leslie Kaelbling, Tomas Lozano-Perez. IROS PR2 Workshop, September 2011. Extended Abstract.
New York Times, Your Very Own Cookie-Baking Robot