Astro utilizes a deep neural network to learn new tasks through a process of trial and errors and could find use in applications such as the detection of guns and explosives, guiding the blind, exploring hazardous environments, or assisting soldiers on the battlefield.
We've already seen several dog-like quadruped robots, which move by walking with their four backward-bending legs. Researchers at Florida Atlantic University are now getting in on the act, with their artificially-intelligent (and dog-headed) Astro robot.
First of all, Astro's style of locomotion is intended to be more than just an attention-getter. Like the other robo-dogs that are on the market or in development, it's able to withstand attempts at knocking it over, plus it's well-suited to traversing rough terrain, such as it might encounter in search-and-rescue operations, or when surveying disaster sites.
What reportedly makes Astro special, however, is the set of Nvidia Jetson TX2 graphics processing units contained within its 3D-printed, Doberman pinscher-inspired head. These give the robot a combined four teraflops of computing power, and with a little help from onboard sensors including a radar imaging module, cameras, and a directional microphone, they allow it to interpret voice commands and make sense of its surroundings.
Additionally, the 100-lb (45-kg) robot utilizes a deep neural network to learn new tasks through a process of trial and error.
Who’s a ‘Good Boy?’ Astro, FAU’s Smart Robodog That’s Who, Florida Atlantic University, 14 Aug 2019