Nvidia Dives Into AI Development, Reveals System Allowing Robots to Learn Human Movements

As artificial intelligence becomes more incorporated to consumer electronics, companies will undoubtedly want to make their mark in this field — and tech fans can count Nvidia to that.

Recently, the company that is known for manufacturing graphics cards for gaming and content creation revealed what they already have in terms of AI development.

Nvidia released a video demonstrating their AI system they call “synthetically trained neural networks” and described as a “deep learning-based system.” The company also showcased how this AI system may help robots or machines to operate based on what they can learn and observe from actual human movements and behavior.

Though the video only showed simple tasks for the robot to do, it was still executed correctly. In the said video, a person is seen arranging several cubes with different colors in a specific order. The robot powered by Nvidia’s deep-learning AI system is on standby at a very close distance to "observe." After disassembling the cubes, the robot then comes to play and did arrange the cubes specifically in the same order as the person previously did.

This AI-focused development was a product of the company's robotics laboratory in Seattle that only started its operation in 2017.

One of the objectives of the research, according to TechCrunch, is for Nvidia to develop a program that will allow robots to work close to humans. This will require robots to, first, be able to detect if a human is within near distance and to achieve that, the machines have to learn human movements and behavior.

Meanwhile, in an interview with VentureBeat, Nvidia lead research scientist Stan Birchfield also noted that even today robots still have limitations in terms of adapting to changes because robots “don’t learn their tasks.” This limitations, according to Birchfield, currently requires an expert on the field every time a robot needs to do carry out a new function or task.

“What we’re interested in doing is making it easier for a non-expert user to teach a robot a new task by simply showing it what to do,” Birchfield added.