FratBot

Making the Rethink Robotics Sawyer platform play beer pong.

Our goal

Create a way to play a fair, interesting game of beer pong with a robot – that is, have the robot locate the cups in beer pong from a vantage point similar to a normal player, choose a position to aim at, calculate how to position itself to throw a ping pong ball to that location, and finally throw the ball using a multi-joint-motion.

IMG_3961muted.mov

Demo of FratBot

Playing a 3-cup game of Beer Pong, by detecting the cups using CV and throwing to their location.

Interesting Challenges

Firstly, to detect the cups to throw to, we must figure out how to find the cups in a noisy environment. Also, as we want to make the robot as human as possible, we position the arm where we don’t get a perfect top-down view of the cups, such that the vision problem remains difficult – we are seeing the table from a 45-degree angle.

Since we are throwing balls in beer pong, we want to make sure we are quite accurate - at about 2 meters away, we should have a margin of error of no more than 2”. Additionally, to throw the balls, we must construct an optimization problem to figure out a trajectory to throw our ball in.

We must also determine how to release the ball - either by polling our position or by time to find a more consistent representation – and use that in our optimization model.

Primary Changes Since Demo

We spent the last week integration testing and improving our vision --> physical location pipeline. Ultimately, we were able to complete multiple full-pipeline beer pong games.

Real World Applications

While robots to throw things could be much better solved with specific appendages like pneumatics, our work can be useful for small-scale businesses who would like to automate portions of their processes without the cost of a specific throwing appendage. For example, to move defective toys to a different manufacturing line.

Additionally, using throwing on robots increases their range past their reachable workspace, which allows for the robot to be smaller and thus cheaper.

Being able to detect cups or other containers reliably without machine learning also provides an easy, out of the box solution for targeting or even general identification purposes.

Finally, as suggested by our wonderful TA Tarun: if robots are the only ones who pass fall rush, it'll deny the creation of more trashy frat boys and generally improve the world.