Most robots can’t pick up most objects most of the time. We aim to change that by using a Baxter robot to collect a corpus of manipulation experiences for one million real-world objects.  The field of object detection and recognition is driven by annotated corpora (e.g., ImageNet) which researchers use to train and test models. These corpora consist of photos taken by a human photographer and may contain many examples of objects, but typically only a single view of each individual object. To bridge this gap, we are using an industrial robot, the Baxter, to automatically collect a database of object models.

If you have a Baxter, you can help! If we had all 300 research Baxters working, we could reach our goal of one million objects in just eleven days. If you are interested in participating in the Million Object Challenge, please sign up below. We are in early alpha testing, and we will contact you when our scanning software is released!

We were recently named finalists in Rethink Robotics video competition and you can see our video here: