More than 650 fluid power manufacturers, users and academics have come here to Aachen, a small western German city, to talk about the future of hydraulics and pneumatics technology, as well as current R&D projects. However, last night, what everyone was talking about was robots.
The closing keynote on Tuesday evening was an all-too-short talk (we could have listened to him all night!) by Aaron Saunders, VP of Engineering of Boston Dynamics. Boston Dynamics is the Massachusetts dynamo of a company that even non-engineers know about, due to the company’s famous YouTube videos that feature bipedal and animal-like robots that jump, run, balance and even do backflips.
Saunders talk focused on building “the world’s most dynamic” humanoid robot—the famed Atlas, which has been through a redesign in recent years.
Saunders noted that Boston Dynamics is a very small company, with only 100 engineers. For the last 15 years, he said that his team has been focusing on basic principles of the mechanics of the locomotion of robotics. Their goal has been on changing people’s idea of what robots can do.
“I’m always asked, ‘What’s the purpose? Are you making any money with this?’ The answer is no,” Saunders said, to laughs from the audience. “This robot’s [purpose] is really to drive innovation inside our group, to push us to understand how to marry controls on complex machines. It is also to create an impression of what robots can do. As we move toward the future, we’re getting closer and closer to when we’re going to turn these things into products.”
People also ask Saunders why they are making robots perform tasks such as making them jump. He explained that it forces his team to face a lot of pragmatic problems. In tasks such as jumping, there is a lot of coordination happening—in the upper body, in the legs and the feet. His videos showed the robot’s hands and arms moving to better stabilize itself, and its legs wobble when it landed on soft ground, far different than laboratory conditions.
This is on purpose, he said. “The other thing we do with a lot of the robots at Boston Dynamics that’s kind of unique is we put them out in the real world. Robots in their history have almost always been in the lab environment.”
“In these environments, the robots have to autonomously navigate the terrain,” he said, as another video showed a robot walking up an uneven set of stone steps in a parklike setting. “The only inputs that this robot is getting right now from the operator are simple joystick commands, like go forward, go left or right, and everything else comes autonomously from the control system.”
Evolution of Atlas
The journey that Boston Dynamics took in getting to this robot did not happen overnight. They started in 2009, literally sawing one of their quadrupeds in half to make an early biped robot, as they worked on a government project that used pneumatics. This robot was tethered for power and cooling.
“In 2012, there was a big competition started in the U.S. to use mobile robots to use in disaster response scenarios, and the government asked them to build 10 robots to give to universities to learn how to access these difficult trends,” he said.
Boston Dynamics used a lot of off-the-shelf components to put this hydraulic robot together, which was a 2-m tall robot that was self-contained and weighed nearly 200kg.
“In 2015, we got the opportunity—when we were acquired by Google—to really look inside and focus on things that we thought were important. We used the opportunity to redesign this humanoid robot from the ground up, and we ended up with a robot that’s very similar. It has all the same strength and range of mobility.”
This newer Atlas model is about 1.5 meters tall and weighs 80 kg. It has an increased strength density to near human levels, is completely power autonomous (running between 30-60 minutes, depending on what it is doing) and has 28 degrees of freedom.
“Cramming 28 active degrees of freedom that all do force control, position control, and high bandwidth into a small machine is actually quite challenging,” Saunders said.
Valves were a problem to source. They found, as their scale got smaller and smaller, and moved down to the human scale, there really weren’t many choices to purchase a high-performance servovalves that they could use to do control. So, they developed their own, which features multiple modes, for traditional servo, braking (negative work) and coasting (chamber to chamber). The valve, he said, has a fast response time and extremely low bypass leakage.
Help from additive manufacturing
3D printing technology has also been key to this version of Atlas.
“When we started the program, I’d read a lot of glossy magazine ads about how 3D printing was here. You could use it, you could print and go. That’s not quite true, but it is a very promising technology and it’s evolving rapidly,” Saunders said.
The robot’s leg makeup was, he said, “probably our biggest undertaking. We learned a lot of lessons … we integrated the structure, the manifold and the fluid routing and actuator cylinders all into one structure.”
“We were able to reduce limb inertia significantly, which is a big deal for a walking robot—most of the power in the system goes to swinging the legs through the air and accelerating and stopping them. You do very little work on the world when you’re a biped and you’re walking—you’re actually very efficient. But you need a lot of power to swing legs, especially when they’re heavy, so this [reduction] was a big deal,” he said.
“We have a saying in our company called the bleeding edge. A lot of people talk about leading edge technology and the leading edge for us is when you’re going too far. The leg was very challenging because there was a lot of stuff integrated into it. Just finding a company to hone an actuator cylinder in a 3D printed material that had never been qualified before is a massive challenge. The number of close processing steps you have to go through as opposed to traditional machining really started to erode some of the benefits. In the end, we still saw that benefit in the inertia, but the effort to get this part out was quite significant,” Saunders said.
Similarly, the HPU uses 3D printing to achieve a lot of efficiencies.
“It’s approaching a kilowatt per kg of density, it’s pretty scalable,” he said. “It sits in the center of the robot. It has everything it needs to collect electrical power and put hydraulic power out … All the homeostasis, sensing, filtration, dump valves, everything we need for the power plant is integrated into a printed part. This lets us wrap everything really tightly around the reservoir—and uses empty space that’s otherwise not used.”
Atlas’ manifold has 18 valves, which service the upper body of the robot.
“This is where we are getting close to a sweet spot in printing,” Saunders said, “so we can make very organic structures and minimize pressure drops—get rid of a lot of excess components. It’s kind of exciting, the things that can be done in printing manifolds.”
But, he reiterated, he wants to see component manufacturers come forward and expand their offerings for uses like this.
“For us, I think one of the big things is the availability of small components. I would love to be able to come to a group like this and find more components on the human scale for mobile applications,” Saunders said. “Developing that valve was really fun, but we’re a robotics company—and we’d like to do more robotics and less component development. So, finding places that work with people to develop these small components on the timescales that are relevant is an area that it’d be great to see more of.”