YouTuber teaches Spot robot dog to pee beer on command
What better way to find use-cases for robots than to make them publicly available, right? Well, that does come with a few trade-offs, including control over how the robot is used. Boston Dynamics is gradually understanding this first hand. In a recent video, YouTuber Michael Reeves was seen attempting to teach the company's Spot robot dog to "piss beer".
The $75,000-worth quadruped was originally designed for industrial surveillance and inspection. However, that didn't stop Reeves from leveraging Spot's programmability. With some unconventional modifications, he was able to transform it into possibly the most sophisticated party companion. After Reeves's modifications, Spot identified cups placed on the ground, moved to position itself above the cup, and pissed beer into the cup.
Reeves outfitted Spot with a two-axis gimbal driven by a few servo motors that "aimed" the beer jet toward the cup once the robot positioned itself above the cup. To detect the cup, he jerry-rigged a cheap surveillance camera running custom code he wrote himself. The beer was dispensed using a solenoid-controlled pressurized tank mounted on the quadruped's back.
Reeves's astonishing genius and the absurdity of the whole creation he calls "Pissbot 9000" are the highlights of the video. However, he remarked that his creation worked as he intended only 35 percent of the time. In the video, Spot would either aim incorrectly or knock over a filled cup or would start walking while still "pissing" beer, eventually slipping like a toddler.
Since Boston Dynamics made Spot a commercially available product, they have had to grin and bear it as people have been trying out a variety of use-cases the company hadn't originally planned. Recently, the French military deployed Spot in combat training exercises. That said, this is the innate beauty of robots: There's nothing a little out-of-the-box thinking, modification, and coding can't fix.
In all seriousness, although Reeves said he's "stupid" and "didn't go to school," he correctly identified that the Raspberry Pi is too slow to run TensorFlow-based computer vision routines to locate the cup. Approaching the problem laterally, he explained that the computer just needs to know where the brightest pixels of the image are, which is an elegant and perfectly functional solution!
