How Boston Dynamics upgraded the Atlas robot — and what's next

How Boston Dynamics upgraded the Atlas robot

In 2021, 60 Minutes visited the offices of robotics company Boston Dynamics and met an early model of its humanoid robot, Atlas. 

It could run, jump and maintain its balance when pushed. But it was bulky, with stiff, mechanical movements. 

Now, Atlas can cartwheel, dance, run with human-like fluidity, twist its arms, head and torso 360 degrees, and pick itself up off of the floor using only its feet. 

"They call it a humanoid, but he stands up in a way no human could possibly stand up," correspondent Bill Whitaker told Overtime. "His limbs can bend in ways ours can't."

Boston Dynamics CEO Robert Playter told Whitaker that Atlas' "superhuman" range of motion is keeping with the company's vision for humanoid robots. 

"We think that's the way you should build robots. Don't limit yourself to what people can do, but actually go beyond," Playter said. 

Whitaker watched demonstrations of the latest Atlas model at Boston Dynamics' headquarters in Waltham, Massachusetts. Rather than turning around to walk in the other direction, Atlas can simply rotate its upper torso 180 degrees. 

"For us to turn around, we have to physically turn around," he told Overtime. "Atlas just pivots on his core."

Boston Dynamics' head of robotics research, Scott Kuindersma, told Whitaker that Atlas doesn't have wires that cross its the joints of the limbs, torso and head, allowing continuous rotation for tasks and easier maintenance of the robot.

"The robot's not really limited in its range of motion," Kuindersma told Whitaker. "One of the reliability issues that you often find in robots is that their wires start to break over time… we don't have any wires that go across those rotating parts anymore."

Another upgrade to the Atlas humanoid robot is its AI brain, powered by Nvidia chips.

Atlas' AI can be trained to do tasks.  One way is through teleoperation, in which a human controls the robot. Using virtual reality gear, the teleoperator trains Atlas to do a specific task, repeating it multiple times until the robot succeeds.

Whitaker watched a teleoperation training session. A Boston Dynamics' machine learning scientist showed Atlas how to stack cups and tie a knot.

Kuindersma told Whitaker robot hands pose a complex engineering problem.

"Human hands are incredible machines that are very versatile. We can do many, many different manipulation tasks with the same hand," Kuindersma said. 

Boston Dynamics' new Atlas has only three digits on each hand, which can swing into different positions or modes.

"They can act as if they were a hand with these three digits, or this digit can swing around and act more like a thumb," Kuindersma said. 

"It allows the robot to have different shaped grasps, to have two-finger opposing grasp to pick up small objects. And then also make its hands very wide, in order to pick up large objects."

Kuindersma said the robot has tactile sensors on its fingers, which provide information to Atlas' neural network so the robot can learn how to manipulate objects with the right amount of pressure.

But Kuindersma said there is still room to improve teleoperation systems.

"Being able to precisely control not only the shape and the motion, but the force of the grippers, is actually an interesting challenge," Kuindersma told Whitaker. 

"I think there's still a lot of opportunity to improve teleoperation systems, so that we can do even more dexterous manipulation tasks with robots."

Whitaker told Overtime, "There is quite a bit of hype around these humanoids right now. Financial institutions predict that we will be living with millions, if not billions, of robots in our future. We're not there yet."

Whitaker asked Boston Dynamics CEO Robert Playter if the humanoid hype was getting ahead of reality. 

"There is definitely a hype cycle right now. Part of that is created by the optimism and enthusiasm we see for the potential," Playter said.

"But while AI, while software, can sort of move ahead at super speeds... these are machines and building reliable machines takes time…  These robots have to be reliable. They have to be affordable. That will take time to deploy."

The video above was produced by Will Croxton. It was edited by Scott Rosann. 

f

We and our partners use cookies to understand how you use our site, improve your experience and serve you personalized content and advertising. Read about how we use cookies in our cookie policy and how you can control them by clicking Manage Settings. By continuing to use this site, you accept these cookies.