If you dump a pile of toy bricks in front of a kid, not much time passes before they figure out how to assemble something fun. The pieces may come in different colors and sizes, but as long as they know how the bricks snap together, children will make miniature versions of just about anything they can imagine – from castles and skyscrapers to spaceships and unicorns.
Autodesk customers also love to make things. But instead of having an army of savvy kids, our customers are often using industrial robots to assemble their products and buildings. As precise and powerful as these machines may be, they’re nowhere near as smart or adaptable as grade-school-aged humans.
But what if robots really could learn to assemble anything in the world in the same way that a child learns how to make an airplane out of colorful plastic bricks?
That is a much more complex challenge than it may seem to us non-roboticists.
If you visit a factory today, you’re likely to find teams of people that spend months or even years programming industrial robots to do just one task, over and over again. Like, say, spot welding in precisely the same location, all day every day, on an automotive assembly line.
The programming process is incredibly tedious, complicated, and frequently error-prone. And since training robots is so time-intensive and the result is a highly inflexible (even if productive) machine, the outcome is a business environment that restricts innovation. If it takes nine months to program your robot to perform a single task, you’re not going to change the design of your product or introduce new technology into your factory – it would simply take too much time and money.
Machine learning could change all of that.
For the past two years, a small team of researchers at Autodesk’s AI Lab on Pier 9 in San Francisco has been working on a project called Brickbot. Led by Mike Haley and Yotto Koga, the project aims to redefine how our manufacturing and construction customers engage and collaborate with robots.
(Fast Company had exclusive access to the project since early last year. You can check out their take on Brickbot here.)
By equipping a couple of industrial robotic arms with cameras and various sensors – then creating neural networks that allow them to intelligently process and respond to all the corresponding data – the Brickbot team has created a robotic system that can infer what’s going on in its environment and then adapt to accomplish an assigned task.
The task? Sort through piles of toy bricks, select the appropriate parts for a given design, then assemble that design by snapping and stacking the bricks together in the correct order and position.
This may seem like child’s play, but imagine what might be possible once Brickbot’s automated assembly technology is built out and scaled up to power the factories and construction sites of tomorrow. These environments will require increased flexibility and adaptability, because as our economies demand more customization of products and buildings, changes to designs will need to be implemented seamlessly, in real time.
“By starting with plastic bricks, we’ve been able to keep the project manageable while still having the freedom to experiment from the design stage all the way to a finished product,” said Koga, a software architect with a PhD in robotics. “Now we’re close to taking the next step. We’re planning to work closely with a manufacturing customer and a construction customer to see how the Brickbot technology can be applied in the real world.”
So, while Brickbot may currently be toying around with kid bricks, the technology’s offspring may someday enable our industrial robot sidekicks to make anything.