So, the CoSTAR project is coming along nicely. We have integrated our perception code with the back-end I have thrown together for the Innovation Award competition, and as a result we have a nice product that lets us flexibly adapt to new tasks.

One of the key elements that differentiates our system from others is the way we include object detection. We can detect objects and update our robot’s model of the world while at the same time allowing non-expert users to author complex plans. More on this some other time. For now, here’s a video of a really simple use case: the robot takes objects off the table one at a time and puts them in a yellow bin.

We can handle objects that are anywhere on the table that the robot can reach.