BakeBot Whips Up Some Cookies

Bookmark and Share

The process of creaming butter and sugar, mixing in a handful of dry ingredients and sprinkling in chocolate chips to make cookies is a recipe simple enough for most humans to follow, but for a robot it is a complex task that requires hours of hierarchical planning by roboticists and real-time planning by the robot.

Graduate student Mario Bollini, a member of Professor Daniela Rus’ Distributed Robotics Lab at CSAIL, has spent the past several months programming the research and development platform PR2 robot, developed by Willow Garage, to bake Chocolate Afghans from scratch.

“My task is to have the PR2 bake cookies all the way from locating the ingredients in front of it on the table to putting the cookie in the oven,” said Bollini.

While the project was originally intended as a simple introductory project, it has turned out to be quite challenging due to all of the nuances involved with programming a robot to follow a lengthy list of tasks, while also employing vision, object detection and executing controlled motions.

To bake the cookies, Bollini first has the PR2 examine the table using a laser scanner and stereo camera to locate the cookie sheet and butter. All of the other ingredients and supplies the PR2 identifies by color and size. The PR2 then follows a hard code of the recipe, from mixing the ingredients to scraping the cookie dough onto the baking sheet and patting it into a large cookie.

Graduate student Jenny Barry, a member of the Learning and Intelligent Systems Group led by Professors Leslie Pack Kaelbling and Tomas Lozano-Perez, has focused on developing a compliant controller for mixing the batter and scrapping out the bowl. The controller was then applied to opening the oven door.

Bollini spent two months recalibrating the PR2 to perfectly mix flour, sugar and butter, as dry beans were initially used in test runs to avoid messes. Now that the PR2 can successfully bake cookies, Bollini is planning to submit his work to an upcoming conference in September.

This project was supported in part by Willow Garage and an NDSEG fellowship.

CSAIL is MIT's largest interdepartmetal laboratory, with 900 members and more than 100 principal investigators coming from eight departments. The lab includes approximately 50 research groups organized into three focus areas: artificial intelligence, systems and theory.

Abby Abazorius, CSAIL