Boston Dynamics’ Spot has been out in the world doing useful stuff (and some other things) for long enough now that it’s high time for a software update packed with more advanced skills and new features.
Spot Release 2.0, launching today, includes improvements to navigation, autonomy, sensing, user programmability, payload integration, communications, and more. Some of that more is an improvement to Spot’s physical capabilities—namely, the robot is better at dealing with slippery surfaces (something Boston Dynamics has always excelled at) and now has a better understanding of stairs, the nemesis of legged robots everywhere.
We’ll take a look at what’s new with Spot, and talk with Boston Dynamics founder Marc Raibert as well as Zack Jackowski, lead robotics engineer on Spot, about some of the highlights of the 2.0 update, how Spot now understands what stairs are, and when we’ll finally be seeing that arm hit commercial production.
Spot Release 2.0: Improved Autonomy and API
The big highlight of Spot Release 2.0 is likely the new and improved autonomous navigation APIs, which (while perhaps not quite as exciting as the stair-climbing skills) are consistent with Boston Dynamics’ focus on getting Spot to do commercially useful tasks. And fundamentally, Spot’s job is to use its robust quadrupedal autonomy to put commercial payloads where they need to be.
“The vision for the robot has always been for it to be a mobility platform,” Zack Jackowski told us. “The thing that Spot does for you is to reliably get a thing you care about to a place you care about, and I think we’re finally getting there with the release of this autonomy API.”
Up to this point, Spot’s autonomy has been at a sort of enhanced teach-and-repeat level, where you drive the robot around the environment that you want it to work in by hand (via a tablet and joystick), and then next time, Spot is able to navigate that specific route autonomously while doing some basic obstacle avoidance and replanning when necessary. This is fine, and does generally set Spot apart from other robotic platforms designed to wander around with your payloads strapped to their backs, but even for users with nonzero robotics experience, it’s not a trivial thing to get setup and running.
“I think people think of us as a mobility company,” says Marc Raibert. “And that’s what we think of ourselves as. But with the recent round of work we’ve been doing on Spot, we’ve broadened our approach to make that mobility more accessible.”
Team CoSTAR, led by NASA JPL, was the winner of the SubT Urban Circuit, a competition organized by DARPA early this year. Among the robots used by the team were two Spots from Boston Dynamics.Photo: Boston Dynamics
There are a handful of new ways in which users can take advantage of Spot’s ability to navigate through, and to some extent understand, the world around it. The previous API, which allowed users to command Spot to move in specific directions for specific distances at specific speeds, has now been vastly expanded to allow Spot to be directed to navigate autonomously through entire buildings. The initial map creation is still done by hand, primarily because of the complexity of the environments that Spot operates in—Spot’s mobility makes autonomous exploration complicated because the robot doesn’t necessarily have a good way of knowing whether it should explore an area just because it can physically get there, and it can physically get to a lot of places.
Once you have that map, though, Spot is now much more autonomous, able to execute high-level navigation commands like “go to this waypoint” rather than having to follow a predetermined path. The robot can also now be programmed through a decision-tree-based system, so that the user can combine navigation, sensing, and action commands into useful behaviors. Instead of Spot just being legs for your payload, the 2.0 update offers several ways to integrate the payload with the intelligence of the robot, leading to a system that’s much more capable.
Spot Gets Mobility Upgrades
The 2.0 release also includes some important mobility upgrades to make Spot more resilient on slippery surfaces and better able to handle stairs, both of which have been issues for Spot in one of its first commercial deployment environments: construction sites.
Construction sites all seem to have these highly polished concrete floors that are super slippery. Boston Dynamics has worked on slipping (or I guess “slip recovery” would be more accurate) for a long, long time—you probably remember this famous video from a decade ago of BigDog on ice. It was (still is, if we’re honest) amazing to watch, but it’s not really what you want your robot to be doing especially if you or something fragile or valuable are anywhere near it.
“It turns out that you can’t just flail your legs in all directions when you’re on a construction site,” Jackowski explains. “When you slip, you need to understand that you’re slipping long before your leg has slipped out from under you.”
Raibert adds: “You need to be calm.” The robot should ideally be able to understand the kind of surface that it’s walking on, and adjust how it moves to minimize slipping as much as possible. This is especially tricky since Spot’s lower legs are largely passive, without motors or sensors in them, but there’s enough proprioception that comes through the legs that the control system can now manage slippery surfaces while keeping flailing to a minimum.
Spot now understands that stairs tend to be regularly spaced, meaning that it can fit a stair model to its mapping and sensor data and have a much better idea of where it’s safe to place its feet.Photos: Bob O’Connor/IEEE Spectrum
Stairs have been a problem for legged robots since there were legged robots. “Stairs are stressful because you don’t get to mess up,” says Jackowski. “You have to be doing the job right from the first step until you’re done, because a misstep is much harder to recover from.”
Jackowski and Raibert explained to us that Boston Dynamics’ robots have generally treated terrain homogeneously, which comes from their history of developing robots for outdoor operations. In other words, Spot would look at a flight of stairs and just see a hill that happens to have ledges and walls at different heights. This has mostly worked, although going up (and especially down) stairs has been consistently tricky for both Spot and its operators.
Spot 2.0 can now identify stairs as stairs rather than weird hills, and apply some basic semantic knowledge to estimate geometric characteristics that make stair climbing much safer and more reliable. For example, Spot now understands that stairs tend to be regularly spaced, meaning that it can fit a stair model to its mapping and sensor data and have a much better idea of where it’s safe to place its feet—and proper foot placement is what will make (or break) a legged robot on stairs.
“On normal terrain, if you don’t put your foot in exactly the place you want to, it’s still usually okay,” Raibert says. “On stairways, there are a lot of constraints. The legs of our robots are not like the legs of animals, and they don’t have that complicated workspace. Stairs just stress them right to the limit. Knowing that there are stairs helps us focus on getting the stepping to be aligned with what’s available.”
Raibert says that the 2.0 improvements to stair handling also help Spot manage some of the absolute worst kinds of stairs for robots, in particular stairs with gratings and open riser stairs that Spot’s sensors find difficult to make out.
And knowing that it’s on stairs can help the robot out in other ways, says Jackowski. “There are a whole bunch of cool things that you can do once you know you’re on a staircase, like adjusting step timing to match the stair rise and run, or understanding that if you see some stuff on a stair, it’s probably not just a differently shaped piece of terrain that you can step on.”
Raibert tells us that so far, the wild Spots have generally been able to do what commercial operators want them to do, quickly integrating into existing workflows without requiring a huge amount of robotics knowledge or the time and effort and stress and failure that you usually get when you try to shoehorn a research platform into a commercial application. The broader goal with Spot Release 2.0 is to move the robot toward a more cohesive platform that’s easier for folks without robotics skills to do something useful with, adding the payload that they need to Spot and then programming the system as a whole.
The 2.0 release also includes important mobility upgrades to make Spot more resilient and better able to navigate and operate on construction sites.Photo: Pomerleau
“We’re really focused on developing a robot that can do routine tasks reliably,” says Raibert. This might not seem like the most exciting goal, especially considering the sorts of things that Boston Dynamics is best known for, but it’s arguably even more difficult and important than doing backflips, suggests Jackowski. “I started at Boston Dynamics doing a lot of the cool flashy dynamic stuff, but the problems involved in making robots work for real in a way that’s actually valuable to someone are just as fun and interesting from a robotics perspective, and in many cases harder.”
Raibert agrees. “Our history is to try and do anything but be boring, but we’re trying to do something new for us—to do boring stuff and have the result be boring, because that’s what people need.” Perhaps noticing our dismay, Raibert hastened to add that Boston Dynamics is also continuing to do exotic stuff, it’s just not the focus for Spot 2.0.
Coming Next: Spot’s Head-Mounted Arm
Looking ahead, Raibert tells us that he expects commercial availability of Spot’s head-mounted arm to happen at some point this year, which’ll enable all kinds of crazy new applications. “The manipulator is going to open a lot of doors,” he says. “And along with the manipulator comes a lot of additional functionality and features, and a lot of that is hard. We’re working on that—the software to back the manipulator up is really cool, some of it with cloud AI to help the arm figure out what it’s supposed to do.”
Boston Dynamics expects commercial availability of Spot’s head-mounted arm to happen at some point this year.Photo: Boston Dynamics
With that kind of incentive to look forward to, I’m sure you want a Spot of your own, because who doesn’t, right? The bad news is that it’s definitely, absolutely not intended to be a home robot any time soon. The good news is that Boston Dynamics really is trying to get as many Spots out in the world doing different things as it can.
“I’m enthusiastic about offering the robot to anybody,” says Raibert. “And part of our goal now is to find out what works and what doesn’t work. We’re trying to not only sell to people who have an established use case that we’re familiar with, like inspecting a refinery or a power station, but we’re also interested in what’s going to come out of the woodwork as these become more available.”
And no, we still don’t know how much a Spot costs. “We’re working on getting the cost down” is all Raibert would tell us.
There is one other way to get your hands on a Spot: Work for Boston Dynamics during a pandemic shutdown. Raibert tells us that there are about 35 robots temporarily living at home with Boston Dynamics engineers to help people stay productive during the pandemic. “I’ve seen demos from people’s homes, and I think some of them may make it to YouTube.” Excellent!
If you’re one of the hundred or so commercial users who have a Spot, the 2.0 update is available immediately. And if you don’t have a Spot, you’ll have to remain sad and lonely, just like the rest of us.