MIT Researchers Build Beer-Delivering Robots

If you have an engineering degree, you might want to skip to this paper published about the beer-fetching robots at MIT. The horror! Fortunately for lazy drinkers everywhere, members of the CSAIL (Computer Science and Artificial Intelligence Laboratory) team at MIT recently presented a system of robots working together in a “bar” scenario that consisted of a PR2 robot “bartender” and two four-wheeled open-source “Turtlebot” kits acting as waiters. They could then reason about which orders were required in each room and when other bots may have already delivered drinks, enabling them to search most efficiently for new orders and deliver the items each of the offices accordingly. Engineers claim that robots teamwork nearly depends of robotic communication, because of the dynamic in real human situations. The sedate human then simply has to reach out and use a single finger to press a single button – the mechanical equivalent of grunting- to order the next round.

A key challenge in making robots work effectively together is the uncertainty factor. And the researchers behind it say their bartender bots could have a number of other useful implications than just helping you get a buzz on. Machines that mix drinks with the help of an app and robot bartenders that are capable of creating cocktails exist but the notable thing about these bartending robots is not their individual skills but their teamwork. Robots deal with several kinds of it, including uncertainty related to outcomes, sensors and communications.

Normally, this type of interaction poses limitations because robots can’t always communicate with each other, and they don’t always have accurate information about their location or the location of things around them. The team described these enhancements as “macro-actions”, their idea being that robots fulfill a general task with various consequences without the need of being assisted all the way. “In the bartender and waiters problem, MDHS was able to automatically generate controllers for a heterogenous robot team that collectively maximized team utility, using only a high-level model of the task”.