Most scientists, with the exception of beleaguered graduate students, don’t exactly have the reputations of party animals. Yet most enjoy a good drink now and then, and some researchers (such as those discussed in this blog’s last post) have devoted their efforts towards developing better beverages. Other scientists are researching how to make the process of getting those drinks easier: Sebastian Loth and his colleagues at Bielefeld University in Germany are developing a robotic bartender by the name of JAMES.
The robot’s name stands for Loth’s project, the Joint Action for Multimodal Embodied Social Systems. As explained on the project’s website, the researchers are exploring how humans combine multiple simultaneous forms of communication, such as “gesture, gaze, body language, facial expressions, and natural-language dialogue,” to get their points across in social settings. “Embodied social systems,” or robots that interact with people, will need to understand this complex communication to fulfill their duties without awkward interfaces, and they could eventually be taught to communicate back at humans in the same way. In a bar setting, for example, a robot bartender might have to distinguish between customers making conversation with other patrons and those trying to order the next round of drinks.
Loth and colleagues reasoned that the best way to determine the rules of such complex communication was to learn from the experts: actual German bartenders. The scientists (presumably while nursing a liter or two) took careful notes on more than 100 orders attempted by patrons, examining which types of communication were most effective at getting a bartender’s attention. As might be expected, two types of actions got the best responses: standing adjacent to and directly facing the bar, which got a patron served within 35 seconds 95 percent of the time, and looking directly at a bartender, which achieved the same result 86 percent of the time. Other behaviors, like talking with other patrons at the bar or examining the drink menu, were unsuccessful; somewhat surprisingly, waving at the bartender was also an ineffective method of getting served.
Although JAMES will have to rely on computer vision to make out these cues, knowing which cues are important will allow it to serve customers more efficiently. This kind of research is a small step towards developing robotic assistants for the sick or elderly, where the consequences of misunderstood communication are more important than momentary thirst. Luckily, the problem of robotically mixing the perfect cocktail has already been solved.