Sick Tricks and Tricky Grips

With a new routine where Atlas assists at a simulated job site, Boston Dynamics engineers have pushed the humanoid robot one step closer to performing real-world manipulation tasks at human speed.

5 minute read

In recent years, Boston Dynamics’ humanoid robot Atlas has wowed the world with its dance moves, and it has hurdled, hopped, and backflipped its way through a complex parkour course.

Now, it’s time for Atlas to pick up a new set of skills. 

In the most recent video from Boston Dynamics, called “Atlas Gets a Grip,” the robot manipulates the world around it: Atlas interacts with objects and modifies the course to reach its goal—delivering a bag of tools to a person waiting at the top of a multi-story scaffold. Atlas grasps, carries, and tosses the tool bag, climbs stairs, jumps between levels, and pushes a large wooden block out of its way before dismounting with an inverted 540-degree flip that project engineers have dubbed the “Sick Trick.” 

At first blush, the routine seems to be a bit of a departure from previous videos, less flashy and more functional. But the new moves represent a natural progression of ongoing research, says Ben Stephens, Atlas controls lead.

“We’re layering on new capabilities,” Stephens says. “Parkour and dancing were interesting examples of pretty extreme locomotion, and now we’re trying to build upon that research to also do meaningful manipulation. It’s important to us that the robot can perform these tasks with a certain amount of human speed. People are very good at these tasks, so that has required some pretty big upgrades to the control software.”

Expanding the Toolkit

To anyone other than a roboticist, the ability to do a backflip (or even to dance a well-executed mashed potato) may seem more impressive than picking up and delivering a bag of tools. But a robot performing manipulation tasks requires a more nuanced understanding of its environment. The dancing in “Do You Love Me?” was entirely blind, without the perception needed for Atlas to respond to its environment. In “Partners in Parkour,” Atlas perceived and moved over and around fixed obstacles (in sometimes breathtaking fashion). 

For this new routine, the robot’s locomotion and sensing capabilities come up against the added challenge of not only detecting, gripping, and moving objects with different sizes, materials, and weights, but also staying balanced while negotiating those objects through the world.

“Parkour forces us to understand the physical limitations of the robot, and dance forces us to think about how precise and dexterous the whole-body motion can be,” says Robin Deits, a software engineer on the Atlas controls team. “Now, manipulation is forcing us to take that information and interpret it in terms of how we can get the hands to do something specific. What’s important about the Atlas project is that we don’t let go of any of those other things we’ve learned.”

For Stephens, one of the most impressive sequences in the new routine is the opening, when Atlas manipulates a large wooden plank. “There’s a lot going on, and it all happens very quickly,” he says. Instead of turning around cautiously, Stephens notes, Atlas performs a 180-degree jump while holding the plank, meaning that the robot’s control system needs to account for the plank’s momentum to avoid toppling over. “And then Atlas places it in exactly the right spot in the world, to use later.”

Pushing the wooden box from the platform is also a deceptively complex task: Atlas needs to generate enough force to cause the box to fall, leaning its weight into the shove without sending its own body off the platform.

Even the flip at the end of the routine is much more difficult than previous acrobatics, because the twist adds asymmetry that doesn’t exist in a regular backflip. Not only is the math more complicated, but in trial runs, Atlas kept getting tangled in its own limbs as it tucked its arms and legs, forcing engineers to troubleshoot and improve the control system so that it was able to choose strategies that avoided self collisions.

“We’re using all of the strength available in almost every single joint on the robot,” Deits says. “That trick is right at the limit of what the robot can do.”

Creating the Future

While some Boston Dynamics robots, such as Spot and Stretch, are commercially available, Atlas is purely a research platform. The Atlas team focuses on pushing the forefront of what’s possible. The leaps and bounds forward in Atlas’ R&D can help improve the hardware and software of these other robots, while also advancing toward a “go anywhere, do anything” robot—capable of performing essentially all the same physical tasks as a person.

Scott Kuindersma, Atlas team lead at Boston Dynamics, acknowledges that robots are unlikely to dip into their bag of gymnastics tricks at job sites. “This is more a demonstration of some of the robot’s new control capabilities, and a fun connection to our prior work,” he says. “Our hope is that, if we can build the foundational technology that allows us to easily create and adapt dynamic behaviors like these, we should be able to leverage it down the road to perform real, physically-demanding jobs with hustle. There are many pieces required to deliver a complete solution in a domain like manufacturing or construction—this video highlights a narrow slice of what we’re working on.”

Stephens says that roboticists are still a “long way off” from creating humanoid robots that can routinely tackle dirty and dangerous jobs in the real world. “Manipulation is a broad category, and we still have a lot of work to do,” he says. “But this gives a sneak peek at where the field is going. This is the future of robotics.”