Dec 12, 2020 - Technology

Helping industrial robots learn to see

Illustration of a robot attempting to see.

Illustration: Aïda Amer/Axios

A startup is bringing improved vision to industrial robots.

Why it matters: As important as the manual abilities of robots are, they need the ability to properly see their environments before they can take a bigger role in industrial workplaces.

What's happening: Akasha Imaging, which focuses on deep imaging and robotic vision, closed on Wednesday a $10.75 million Series A funding round led by Khosla Ventures.

  • Akasha uses deep learning to produce 3D images in unstructured environments, which can help train industrial robots to "see" their surroundings and more precisely handle products of all shapes and sizes.
  • "We can create synthetic data and use that to test [robots] in the real world," says Kartik Venkataraman, Akasha Imaging's CEO. "And because the data is synthetic, we can do our testing remotely."

Details: Vision — as you'll know if you've ever tried to, say, wrap a holiday present after misplacing your glasses — is key to manual work. That's especially true if you're operating in the less than fully predictable real world.

  • With robots, "even if our colleagues on the electronics side have done a perfect job, without good vision it's the equivalent of blindfolded humans trying to do packaging," says Achuta Kadambi, the company's chief scientist. "We help take the blindfold off."

The bottom line: Kadambi believes robotic vision will mean robots eventually "will be able to do tasks that are impossible today" — even for humans.

Go deeper: Robots vs. retail workers

Go deeper