By [ad_1]
I don’t know what’s more impressive, the fact that a bot named Susan dominates at cornhole, or that it can also fabricate its own snazzy looking cornhole boards using power tools (it just needs human assistance with the paint job).
What the hell is cornhole, anyway? Well, it’s what I watch on TV when there’s nothing else on—it’s not unusual for ESPN to air cornhole tournaments, and it’s oddly mesmerizing. It’s also a lawn game. You plop two inclined boards precisely 27 feet apart (or just eyeball it), each with a circular hole cut out towards the top, and then take turns tossing bean bags at the holes. Being drunk is optional.
There are rules and a point scoring system. What it essentially boils down to is you want to get more bean bags into the hole than the person (or robot, incidentally) you are playing against, because it affords the most points (3 points per bag). Sounds easy, but it does actually take a fair amount of skill, like any sport.
David Niewinsky of the YouTube channel Dave’s Armoury built a robotic arm that can “throw a perfect game of cornhole,” and won Nvidia’s “Jetson Project of the Month” for the effort.
Jetson is the name of an AI platform offered by Nvidia for edge computing. It consists of a complete system on module (SOM) with a CPU, GPU, memory, power management, high speed interfaces, sensors, and other bits.
David paired a Jetson kit with a Kuka K20 robot named Susan. It uses a 1080p webcam to ‘see’ the world, and in this case, to identify a cutout with a red painted border in the cornhole board. Susan then calculates the proper trajectory and motion needed by its attached extrusion bar, which serves as its throwing arm, for a perfect beanbag toss.
That means there is a software element at play as well.
“For the software, Dave used several OpenCV functions such as inRange to pick out the red hole from the scene, and findContours to establish the ring around the hole,” Nvidia explains. “Using the relative positions of the camera and the center of the hole, the angle and power for the throw are calculated on Jetson. Lastly, Jetson communicates these calculations to Susan through the network via the KUKA.ethernetKRL software package.”
Precise beanbag tosses are one thing, but could it beat a professional player who might alter their strategy to block part of the opening with an imperfect throw? I don’t have the answer, but Susan and a human teammate did beat a pair of competitors from Hacksmith Industries rather easily. Being a basketball fan, I’d have called her Suzy Buckets, because it’s just one score after the next.
This is the fun side of AI. Or depressing, if you’re tired of bots beating us flesh and blood mortals at things like Dota 2, checkers, and other games. But whatever—congrats to Suzy Buckets for another win for team AI.
[ad_2]
