The marketing specialists at Boston Dynamics produced two major robotics news cycles last week. The largest of the two was, naturally, the announcement of the electric Atlas. As I write this, the sub-40-second video is steadily approaching five million views. A day earlier, the company touched the hearts of the community when it announced that the original hydraulic Atlas was being put out to pasture, a decade after its introduction.
The accompanying video was a celebration of the ancient Atlas’ journey from the DARPA research project to an impressively agile bipedal robot. However, a minute later, the tone changes. Ultimately, “Farewell to Atlas” is both a celebration and a tale of mistakes. It’s a welcome reminder that every time the robot makes the video landing, there are dozens of slips, falls and sputters.
Image credits: Boston Dynamics
I have long advocated for this type of transparency. It’s the kind of thing I’d like to see more of in the world of robotics. Simply showing the highlight reel does a disservice to the effort put into getting those shots. In many cases, we’re talking years of trial and error put into getting robots to look good on camera. When you only share positive results, you are setting unrealistic expectations. Bipedal robots fall down. In that sense, at least, they are like us. Like agility put it on recently“Everyone falls sometimes, what defines us is the way we get up.” I would go a step further and add that learning to be liked is equally important.
The company’s newly appointed CTO, Pras Velagapudi, recently told me that seeing robots fall into work at this stage is actually a good thing. “When a robot is actually out in the world doing real things, unexpected things will happen,” he notes. “You’ll see some falls, but that’s part of learning to run for a long time in real-world environments. “It is expected and it is a sign that you are not setting things up.”
A Quick Look at Harvard Rules for falling without injury reflects what we intuitively understand about falling as humans:
- Protect your head
- Use your weight to direct your fall.
- bend your knees
- Avoid taking other people with you
As for robots, this IEEE Spectrum piece from last year It’s a great place to start.
“We’re not afraid of a fall; we don’t treat robots like they’re going to break all the time,” Boston Dynamics CTO Aaron Saunders told the publication last year. “Our robot falls a lot and one of the things we decided a long time ago [is] that we needed to build robots that could fall without breaking. If you can go through that cycle of pushing your robot to failure, studying the failure, and solving it, you can move forward to where it doesn’t fall. But if you build a machine or a control system or a culture around never falling, then you will never learn what you need to learn to keep your robot from falling. We celebrate falls, even those that break the robot.”
Image credits: Boston Dynamics
The topic of crashes also came up when I spoke with Boston Dynamics CEO Robert Playter ahead of the launch of the electric Atlas. Notably, the short video begins with the robot in a prone position. The way the robot’s legs arch is quite novel, allowing the system to stand up from a completely flat position. At first glance, it almost seems as if the company is showing off, using the eye-catching movement simply as a method of showing off the extremely robust custom-made actuators.
“This will have very practical uses,” Playter told me. “The robots are going to fall. You better be able to get up from your stomach.” He adds that the ability to rise from a prone position can also be useful for carrying.
Much of Boston Dynamics’ learnings about crashes came from Spot. While there is generally more stability in the quadruped form factor (as evidenced by decades of attempts and failures to kick robots in videos), there are simply many more hours of Spot robots working in real-world conditions.
Image credits: Agility Robotics
“Spot travels about 70,000 kilometers a year in the factories and carries out about 100,000 inspections a month,” adds Playter. “Eventually they fall. You have to be able to get back up. Hopefully it reduces your fall rate, we have. I think we fall once every 100-200 km. The drop rate has really gotten small, but it happens.”
Playter adds that the company has a long history of being “tough” on its robots. “They fall and they have to be able to survive. “The fingers can’t fall off.”
Looking at the Atlas outtakes above, it’s hard not to project a bit of human empathy onto the robot. He actually appears to fall like a human, bringing his limbs as close to his body as possible to protect them from further injury.
When Agility added arms to Digit back in 2019, there was talk of the role they play in falls. “For us, arms are simultaneously a tool for moving through the world (think getting up after a fall, waving your arms for balance, or opening a door) and at the same time useful for manipulating or transporting objects,” he said. co-founder Jonathan. Hurst noted at that time.
I spoke to Agility a bit about the topic at Modex earlier this year. Video of a Digit robot falling to the floor of a convention a year earlier had circulated on social media. “With a 99% success rate over about 20 hours of live demos, Digit still suffered a couple of crashes on ProMat,” Agility noted at the time. “We don’t have proof, but we believe our sales team orchestrated it to be able to talk about the durability and quick-change limbs of the Digits.”
As with the Atlas video, the company told me that something resembling a fetal position is useful to protect the robot’s legs and arms.
The company has been using reinforcement learning to help downed robots recover. Agility stopped Digit’s obstacle avoidance in the video above to force a fall. In the video, the robot uses its arms to mitigate the fall as much as possible. He then uses his reinforcement learning to return to a familiar position from which he is able to rise again with a robotic push-up.
One of the main selling points of humanoid robots is their ability to integrate into existing workflows: these factories and warehouses are known as “brownfield”, meaning they were not custom-built for automation. In many existing cases of factory automation, errors mean that the system is effectively shut down until a human intervenes.
“Rescuing a humanoid robot will not be a trivial task,” says Playter, noting that these systems are heavy and can be difficult to correct manually. “How are you going to do that if it can’t take off on its own?”
If these systems are truly going to ensure uninterrupted automation, they will have to fall down and get back up.
“Every time Digit goes down, we learn something new,” adds Velagapudi. “When it comes to bipedal robotics, falls are a wonderful teacher.”
