Watch And Learn: Robot Picks Up New Skills In Seconds

A breakthrough technology means robots can literally “watch and learn” without the need for any programming.

Just perform an action – packing items or stacking containers, for example – and they will observe, analyze and copy, often in a matter of seconds.

A team of programmers and engineers could take days or weeks to do the same thing. And if the task needed altering, they’d have to come back again.

The software, developed by Israeli startup Deep Learning Robotics (DLR), can be downloaded onto any computer, which serves as the robot’s controller.

Users simply select an on-screen option to start the “lesson”, prompting the robotic arm to watch and learn. 

The tech observes the person doing the task, extracts the necessary data, then converts it into a simple universal language so the robot can mimic the task. 

“It all happens in a matter of seconds, sometimes minutes at the most,” says Carlos Benaim, Founder and CEO of the company. 

He says no other company has managed what they’ve managed – to teach robots by having them watch people.

There’s a lot of research and development, and he’s sure others will follow, but so far, he says DLR is the first to reach production lines. 

The software connects to any camera, giving the robotic arm ‘eyes’ so they can watch the action. It retrieves the raw data of the 3D environment it is observing, and processes it so the robot differentiates between the person carrying out the task, the objects that are being handled, and the purpose of the task.

For example, when it sees a person take an object and put it in a different location, it understands that moving the object is the goal of this task. Once it learns the task, it performs it perfectly every time, and doesn’t ever get tired.  

Users can also edit the task through the software after the robot has learned it. “After it learns the task, you can always change some parameters that will adapt it more to what you want it to do,” he says.

That said, the accuracy of the task the robot learns depends on the camera and the type of robotic arm that is being used. 

“Our whole goal is to make automation through robotics so easy that the people that still work manually can do it with a robotic arm,” Benaim tells NoCamels.

“It can take months to implement a robotic arm in factories. And the moment you need to change something you need to bring in engineers again. We eliminate this barrier.”

The tasks that the robots can learn seem limitless. Benaim shares that at one point, during the company’s collaboration with the University of Haifa, a student taught a robotic arm to identify a picture and replicate the drawing all by itself.

But the technology is mostly used for industrial purposes – like packing, logistics (like automating online supermarket orders), assembling technology like laptops, painting, polishing, welding, and soldering.

One of its clients is a potato chips producer, and has taught its robots to sort potatoes – normally a manual task – removing any that are too big or too small from the processing line. 

Right now, the technology is used solely for robotic arms – as opposed to AI-powered robot assistants, for example. But Benaim says the sky’s the limit.

“I think at the end of the day, this type of technology will be able to be used at home. There are so many tasks that we do manually today that we would rather have a robot do. And eventually this will be possible.”

But how does it connect to any kind of robotic arm, when there are countless robotics manufacturers? It’s like Windows software being used in almost every computer, no matter the type, says Benaim. 

Each manufacturer has an open connection – or an open door for external connections, like a network port. The software is able to translate the smart tasks that are detected into a low-level language that every robot understands. 

Benaim says robotic arms will not replace people’s jobs, and if anything, they increase employee productivity because they don’t need to do the same manual task over and over again. 

“I think that we’ve seen many companies whose robots are underused. They’re only used for one particular task. And in many cases, it’s related to the fact that they are used for what traditional technologies allow them to do.

“I believe that adding our technology would expand the type of tasks and the amount of tasks that the same robots could do.”

And as for technologies that make robots easier to use, like cobots, which are robotic arms that work side-by-side with humans, Benaim doesn’t see them as competitors, but as complements to his company’s technology. 

DLR’s clients include leading robotic arms manufacturers in the Far East, Fortune 500 companies, and household names like Panasonic and PepsiCo.

Subscribe to NoCamels weekly newsletter and get our top stories