“Navigating via Nvidia’s ‘Cosmos’: How AI Guides Humanoid Robots Through the Planetary Maze”
“Nvidia’s ‘Cosmos’ AI Helps Humanoid Robots Navigate the World”
“When the founders of a startup that uses artificial intelligence to teach robots how to pick up objects, many of which have unusual shapes, decided to test their system, they turned to a seemingly nonsensical way to make things harder: They put on silly costumes and started dancing around.”
In a world where we find machine intelligence more accessible and increasingly indispensable daily, Nvidia comes up with an AI that operates like a toddler playing with Legos. Except in this case, the toddler wears a techno-infused spectacle and dances around to train robots. Absurd, isn’t it? But, if you expected less dramatic methods to test an AI system, you’re not paying enough attention.
Nvidia’s new AI, nicknamed CoSMo, is designed in a way that it teaches robots to recognize objects and pick them up, regardless of their shapes. This might sound like a simple kindergarten game, but let’s imagine a robot-trained toddler playing with a box-shaped toy, a round one, and then a star-shaped one – all while watching dancing adults. The CoSMo is expected to roll with these punches pretty similar to the poor confused toddler.
Talking about the science stuff behind it, CoSMo utilizes videos from eight different cameras, ultimately creating distinct scenes that involve objects moving around semi-randomly. The simultaneous localization and mapping (SLAM) combined with convolutional neural networks (ConvNets) help the AI to understand these scenarios, making things easier for the four-wheeled toddler.
Jennifer Granick, surveillance and cybersecurity counsel for the ACLU Speech, Privacy, and Technology Project, rightly questioned, “Shouldn’t a self-driving car be able to recognize that something novel is moving around in front of it?”. Well, Jennifer, if we live in a world where humans can’t seem to figure out the complexities of self-assembly furniture, can we truly blame an AI that’s trying to make sense of dancing humans dressed in spandex?
So, while it’s amusing to think about how these founders decided on the testing method – “Hey, let’s dress up in these ridiculous outfits and prance around to confuse our AI toddler!”, we’ll just have to wait and see if their unusual method proves to be effective. In the meantime, here’s to Nvidia’s attempt to train the self-driving industry’s new toddlers. If nothing else, it certainly adds a humorous spin to the monotonous image of a tech-driven future.