For the first time today, both Atlas robots have completed the complex obstacle course flawlessly. Or, almost flawlessly.

The first of the two robots ran up a series of banked plywood panels, broad jumped a gap, and ran up and down stairs in the course set up on the second floor of the Boston Dynamics headquarters. The second robot leapt onto a balance beam and followed the same steps in reverse, and then the first robot vaulted over the beam. Both landed two perfectly synchronized backflips, and the video team has captured every move. 

And yet, the robotics engineers who have been working on this routine for months barely take time to celebrate. Moments after the cameras cut they’re huddled together, making changes before the next take. Although this most recent attempt was nearly perfect, it was not precisely perfect, not quite. After the robots completed their backflips, one was supposed to pump its arm like a big-league pitcher after a game-ending strikeout – a move that the Atlas team calls the “Cha-Ching.” 

The robot did pump its arm, but it also stumbled a bit on this simple move. It was just the slightest stutter step, something most people watching the video would never notice. But the Atlas team notices every detail and they want to get it right. 

“We hadn’t run that behavior after the backflip before today, so that was really an experiment,” says Scott Kuindersma, the Atlas team lead at Boston Dynamics. “If you watch the video closely, it looks a little awkward. We’re going to swap in a behavior we’ve tested before so we have some confidence it will work.”

The perfectionism on display here is admirable, and one gets the sense that this sort of attention to detail is what has produced robots that are capable of completing such a futuristic, impossible-seeming routine. But it also raises a question that at first seems surprising, but then seems obvious: Why does it matter? 

Why does it matter if Atlas stutters when it pumps its fist? Why does it matter if it pumps its fist at all, for that matter? And while we’re asking questions, what exactly is the use of running the robots through an obstacle course? Why the banked plywood panels, why the vault, why the backflips? 

In short: Why parkour? 

Building a Foundation

A robot’s ability to complete a backflip may never prove useful in a commercial setting. (Atlas is a research platform, not a commercial product.) But it doesn’t take a great deal of imagination or sector-specific knowledge to see why it would be helpful for Atlas to be able to perform the same range of movements and physical tasks as humans. If robots can eventually respond to their environments with the same level of dexterity as the average adult human, the range of potential applications will be practically limitless. 

“Humanoids are interesting from a couple perspectives,” Kuindersma says. “First, they capture our vision of a go-anywhere, do-anything robot of the future. They may not be the best design for any particular task, but if you wanted to build one platform that could perform a wide variety of physical tasks, we already know that a human form factor is capable of doing that.”

“From a technical perspective, humanoids present several challenges that we welcome as a research team,” he adds. “Their combination of size and complexity creates hardware design tradeoffs related to strength to weight ratio, runtime, range of motion, and physical robustness. At the same time, our control team has to create algorithms that can reason about the physical complexity of these machines to create a broad set of high energy and coordinated behavior. Ultimately, pushing the limits on a humanoid robot like Atlas drives hardware and software innovation that translates to all of our robots at Boston Dynamics.”

Ultimately, pushing the limits on a humanoid robot like Atlas drives hardware and software innovation that translates to all of our robots at Boston Dynamics.

Scott Kuindersma, Atlas Team Lead

Parkour, as narrow and specific as it may seem, gives the Atlas team a perfect sandbox to experiment with new behaviors. It’s a whole-body activity that requires Atlas to maintain its balance in different situations and seamlessly switch between one behavior and another. 

“It’s really about creating behaviors at the limits of the robot’s capabilities and getting them all to work together in a flexible control system,” Kuindersma says. “There are many important problems that parkour doesn’t force you to address, but that’s not the point. We’re not trying to solve everything all at once. The work we’re doing now is allowing us to create a solid foundation for tackling the next set of research problems.” 

The Evolution of Atlas

Looking back over five years of Atlas videos, it’s easy to lose sight of just how much progress the team has made during that time. In fact, some of the improvements are invisible to the average viewer, even though they represent giant leaps (quite literally, in this instance) in technology. Although Atlas was doing dive rolls and handstands and backflips in earlier videos, the underlying processes for controlling those moves have evolved. 

“Atlas’s moves are driven by perception now, and they weren’t back then,” Kuindersma explains. “For example, the previous floor routine and dance videos were about capturing our ability to create a variety of dynamic moves and chain them together into a routine that we could run over and over again. In that case, the robot’s control system still has to make lots of critical adjustments on the fly to maintain balance and posture goals, but the robot was not sensing and reacting to its environment.”

In this iteration of parkour, the robot is adapting behaviors in its repertoire based on what it sees. This means the engineers don’t need to pre-program jumping motions for all possible platforms and gaps the robot might encounter. Instead, the team creates a smaller number of template behaviors that can be matched to the environment and executed online.

“We decided to add the banked turn pretty late in our development process” says Yeuhi Abe, a senior control engineer on the Atlas team. “We were able to leverage tools developed for creating jog motions in other contexts to quickly create a prototype that we refined using a combination of simulation and robot testing.”

Simulation is an essential development tool for the Atlas controls team, both for evaluating new behaviors prior to robot testing and for ensuring that new software changes don’t negatively impact existing capabilities. But there’s still no replacement for hardware testing, particularly in performance-limiting motions like vaulting.   

About that vault: unlike high-flipping gymnastics vaults, a parkour vault is a slightly less flashy method designed to get a runner over a low wall or obstacle – in this case the balance beam, only a few feet high. Atlas places its arm on the beam and then hoists its body over the structure. For many humans, this sort of vault would be relatively easy (especially in comparison to a backflip), but for the Atlas team, it represented a formidable new challenge. 

“If you or I were to vault over a barrier, we would take advantage of certain properties of our bodies that would not translate to the robot,” Kuindersma notes. “For example, the robot has no spine or shoulder blades, so it doesn’t have the same range of motion that you or I do. The robot also has a heavy torso and comparatively weak arm joints. Extending our tools to help us find solutions that worked within these constraints was what made the vault an interesting challenge.”

Joys & Frustrations

During filming, Atlas gets the vault right about half of the time. (A natural consequence of pushing robots to their limit is that, sometimes, those limits are met.) On the other runs, Atlas makes it over the barrier, but loses its balance and falls backward, and the engineers look to the logs to see if they can find opportunities for on-the-fly adjustments. 

“There are a lot of pretty exciting behaviors here, and some of them are not totally reliable yet,” says Ben Stephens, the Atlas controls lead. “Every behavior here has a small chance of failure. It’s almost 90 seconds of continuous jumping, jogging, turning, vaulting, and flipping, so those probabilities add up.”

Stephens adds that this is the first time Boston Dynamics has filmed two robots performing parkour together. “We had never actually done the two robots together until two weeks ago,” he says on the day the routine is being filmed. “We’re in a place now where it should work. We think we’ve caught all of the major failures, and now it’s just down to those small probabilities.” 

In the end, as always, everything comes together. The video crew gets a take that everybody is happy with – down to the final celebration moves. The work is painstaking and incremental, but when both robots make it to the end of the routine, the engineers have to stop themselves from cheering before the cameras stop rolling. 

“I think that’s one of the joys of robotics, that we’re solving really hard problems, and with that comes the inevitable frustrations along the way,” Kuindersma says. “I find it hard to imagine a world 20 years from now where there aren’t capable mobile robots that move with grace, reliability, and work alongside humans to enrich our lives. But we’re still in the early days of creating that future. I hope that demonstrations like this provide a small glimpse of what’s possible.”