What makes this work is the motors.[1] It's using 3-phase drone prop motors with no gear reduction. With direct drive, the mechanics of the joints are simple. Not too expensive, either. Each motor is about US$120.
I used to work on legged running back in the 1990s. But there were no motors good enough to build a small machine like this back then. All we had were R/C servos, gearboxes, and screw drive actuators. So I was stuck in simulation. All the early legged robots were mechanical nightmares. Springs, cables, pulleys, worm gears, harmonic drives... Now, at last, enough torque in a small package.
What you want in a leg muscle is well known - something that behaves like a spring/damper combo for which you can set the spring constant, the damping constant, and the spring's zero point. People have been trying to do that by mechanical means for decades. There were pneumatic systems, hydraulic systems, things with springs, pulleys, and wires, and a kludge called a "series elastic actuator", which is a stiff spring-damper combo in series with a screw drive actuator. They're all terrible.
"You cannot strip the teeth of a magnetic field" - GE electric locomotive sales rep, early 20th century.[2] That's why they can drop this thing and have it spring back. Try that with any of the mechanical nightmares and the results will not be good.
I was involved in building --or more accurately, rebuilding-- one. I went as far as doing a whole new design CNC-machined out of aluminum.
One of the problems with this design is that these drone motors are not designed for static operation drawing lots of current without any cooling whatsoever. The 3D printed plastic parts are pretty much insulators. We had one motor smoking. The thing could not stand due to its own weight. This (and other factors) led to thinking of a new design using similar principles, a better motor and an aluminum structure that could be used to move some of the heat away from the motors.
I was wondering about that. The big innovation from Schaft was that they liquid-cooled their motors.
That allowed them to build a full size humanoid without going hydraulic.
What happened that made electric motors so much better since the 90’s? It’s not like we made groundbreaking discoveries about magnetism, right?
Neodymium-cobalt magnets and 3-phase drive in small packages. Also, a big market for high-power low-weight drone motors brought the price down.
That's a nice little device. Those might make good finger actuators.
Linear motors have long been a tiny niche. I'd looked at Aura linear motors, which was the leading brand back in the 1990s, but Aura got themselves into serious legal trouble. They're back now, but mostly as an aerospace supplier. Power to weight ratio in small linear motors has seldom been good. It's not clear whether this is fundamental or just lack of engineering effort due to small demand.
Thanks, that's great info. Am I mistaken in remembering that the video shows the thing standing? Is it actually not capable of this? Or is there some very short time-limit?
I am sure it is. However, the design uses very expensive custom boards to save a lot of weight. The build I was involved with used Odrives and that was enough to have issues. Pre-stretching the belts is also important.
If I had to guess I would suggest that, while 3D printing is a great enabling technology one of the issues it suffers from is that of tolerances. I'll preface this by saying I did not take any measurements at all on the parts as printed. If I did, I would not be surprised to discover issues that, in the aggregate, could lead to performance issues.
The one I was involved with could never jump like that. My guess is that umbilical, at a minimum, carries power. Based on what I know, I cannot imagine the robot doing what is shown on these videos while loaded with the two battery packs that are part of the design.
Please keep in mind that this is a single data point. I am sure there have been many successful builds. It's a nice design. I was not intimately involved in the original build of the robot we dealt with. I was involved in a complete tear-down and reconfiguration to make it work. As I mentioned, this led to the design of a lightweight quadruped that is fully CNC machined and uses more powerful and capable customized drone motors.
I'd be curious to learn more about the motor selection for the complete tear down and rebuild. One of the biggest buys of the Solo (in the article) design is the 'high bandwidth' current/torque control, and the price/performance seemed (at the time) to be pretty great.
> It's using 3-phase drone prop motors with no gear reduction.
If I'm understanding correctly, this is not accurate. The paper and the Github repository describe it as a "dual stage timing belt transmission with a 3:1 gear reduction on each stage".
Aw, you're right. We're still not there on tiny direct drive. I've seen direct drive robots with large "pancake" motors, and thought they were driving those little guys that way. But no. They still need a lot of reduction. I'm surprised that they bounce as well as they do while back-driving through 9:1
I bought a couple 'moteus' motor controllers from Josh Pieper at mjbots to build a small two axis gimbal to point an antenna. It was my first foray into any kind of hardware project of this nature and I honestly didn't fully respect the amount of force these small brushless motors can generate. My head is thinking 'these things run on batteries and meant to spin at 3000 rpm, how much torque can they have?'
A lot.
I've broken a half dozen mounts, ripped a circuit board in half and smoked a few passives in the process. My recommendation is to start with a small motor when you're tinkering with this stuff haha.
(I'm no expert but super happy customer of mjbots btw)
I'm not sure any of the big legged robots are using direct drive actuation yet; MIT Cheetah and Cheetah mini use quasi-direct drives with a sub-10:1 gear ratio for super high proprioceptive sensing. ANYmal actually uses series-elastic actuators, since they allow the actuator to operate at a higher speed and lower output torque, thus making them more efficient than a quasi-direct drive setup found in other quadrupeds [1]. They also allow for some energy reuse, since they can release energy stored by compressing the spring. The downside is, like you said, reduced control bandwidth, since the spring element will behave optimally under a given set of conditions and, going outside that operating range, will either be more stiff or less than is ideal.
That said, for the kind of work ANYmal is currently being deployed on (site inspection mainly, I believe), they're hardly doing hyper-dynamic movements, so I think selecting SEAs over quasi-direct drive was a fairly judicious choice. Just like it makes more sense to have quasi-direct drives in a highly dynamic robot like the MIT Cheetah; it depends on the use case.
What do you think about these direct linear actuators [0]? They seem super simple, and like a closer match to the way biological muscles work. Is there some reason why rotary (?) actuators are the default choice for robotics?
Edit, also what happened that made electric motors so much better since the 90’s? It’s not like we made groundbreaking discoveries about magnetism, right?
Well those are voice coil motors, similar to a subwoofer driver in construction. I don't know how efficient those are, but I did release some info on a brushless linear motor I designed. My linear motor design is like a regular rotary brushless motor but rolled out and made tubular. Maybe better power density than voice coil motors, tho I am not sure. The motor worked great, but I got busy with other things and never commercialized it. I have some details on github.
I can't give an answer which would pass a proper roboticist without them wrinkling their nose, but my guess is that rotary actuators are more simple and space efficient to implement, and can generate the necessary high torques for these systems.
When using a linear actuator, you need to offset it from the joint so that it generates torque (that's part of the reason we have kneecaps, actually; it increases the distance between the centre of rotation of our knee and where our leg muscles are actually pulling from, increasing the output torque). This involves more mechanical complexity in the design compared to rotary actuators, which can just be slapped onto the joint directly. That offset also takes up space, so your system is now less compact.
None of the heavier legged robots I know of use direct drive actuators; ANYmal uses Series Elastic Actuators, MIT Cheetah and (maybe) Boston Dynamics Spot use Quasi-Direct Drive actuators with a low gear ratio gearbox (sub-10:1, I believe), RealHyQ uses hydraulic actuators. Electric actuators simply can't output the amount of torque these systems need to carry themselves around, so they need some form of gearing.
The linked actuators you shared are direct driven, so they would need some gearing on their output. I had trouble finding anything about gearing direct drive linear actuators, which leads me to believe it's not really a thing, because if you need high loads, you'll jump to geared linear actuators, which are really just rotary actuators with a belt or rack and pinion!
Aside from the mechanical challenges associated with using a linear actuator setup, it's (to my knowledge) not necessarily better to follow biology when it comes to robots.
Wheeled robots can move faster on flat terrain than any animal, so legs aren't ideal when you have access to gas-powered engines or electric actuators. Airplanes with turbines can carry much heavier loads than birds because flapping wings just doesn't scale up to the weight of a jumbo jet with luggage and passengers. Legged robots are actually more energy efficient when they don't try mimicking an animal's gait because they don't have organic muscles arranged in the same way. Any finally, rotary actuators are a better choice than linear actuators (for now) because we don't have hyper-compact, super torque-dense linear actuators that can match our muscles.
This isn't to say that biologically inspired designs can't be better, and there's a lot of research into finding ways of replicating the natural properties of certain materials, such as the strength of silk or hardness of the shells of some crustaceans. But in the end, with the technology we have, rotary actuators provide better torque density, and a more compact and simple design than linear actuators.
[1]: I didn't cite anything here because of time constraints, but it's an amalgamation of half-forgotten knowledge from some research I've done on the topic of legged robots
How does this relate to MIT's mini-cheetah(1)? A quick search didn't turn up any plans/code for the cheetah except some simulator software(2). Is this an open source 'work-alike' ?
I've read that, but am not convinced. It's very dependent on the particulars of the situation, the goals, the acceptable losses, and the belligerent forces.
A couple of situations:
1) Tyrannical government or Upstart AI wants to destroy dug in and dispersed local militias. Government will use existing military delivery technologies. AI will use non-human elements so blowback is not a concern. It seems here that gas is cheaper than bombs or bullets.
2) Upstart State or AI wants to do a surprise attack to depopulate or logisticall overwhelm a strategic city. Civilian casualties are desired. Again, gas seems like a clear win, here.
The idea that militaries have NBC protection that will just negate chem/bio warfare is a bit of a dream. The operational friction from having to use all that protective gear and protocols would be huge. Even just little things like not being able to aim properly through a rifle scope depending on which gas mask you're wearing. Also don't underestimate the fear/terror/shock factor of being on the receiving end of this horrible stuff.
The future isn't pretty, unless we can figure out a way to settle down and coexist.
There are also important questions about warning time and strategic responses.
Let's say an upstart AI or group decide to wipe out humanity, or at least a country or continent. What is the time from initial detection to the realization that we as humans are at the end of the world? What do we do then? Do we nuke someone? Something? Everyone?
Any upstart taking over the world strategy is going to have to grapple with this. Chemical weapons have immediate effects. Bio-weapons will take time to spread - time during which they could be detected and possibly mitigated.
I'm sure a lot of actors have taken note of COVID-19's asymptomatic transmission. I'm sure people are thinking of bio binary weapons. Step 1) Silently infect the target population. Step 2) Flip the kill switch and they all die simultaneously.
I dunno. Chemical plant seems relatively crude and cheap compared to the equipment needed to design and build a reliable, effective, and controllable bio-weapon. Two-photon microscopes and nanopore gene sequencing aren't cheap. If it doesn't have to be reliable or particularly effective, and especially not targeted, then maybe you're right. What am I missing?
Quick/friendly reminder that this is a dangerous conversation to be having in public -- there is no telling what will happen to old comments in a few years (or decades,) when the disruption is further along.
Quite serious. History tends to happen when you least expect it to.
A former employer once forwarded me a deep-dive article about a guy who was archiving old '90s usenet posts. He explained the process by reference to a sample post, chosen at random, from the whole of the Usenet, '91-'99.
That guy's sample-post? A one-off I'd made, complaining bitterly about JDK licensure or something, from around 1998, when I was a loudmouth teenage Stallmanite.
It now occasionally comes up for discussion during job interviews. It came up when I applied for a work visa to USA.
Stay safe, and don't let your next job interview turn to the topic of [checks notes] resource-efficient weapons of mass destruction.
I have an tangential question about open source projects and names. "Boston Dynamics" is a trademark of Boston Dynamics Inc. Can using "Dynamic" like this project does cause legal problems, considering the name can be an allusion to the trademark?
And I have a set of Boston Acoustics speakers sitting on my desk here that are almost certainly older than the Boston Dynamics trademark.
Hmmm, just looked it up: "Boston Dynamics is an American engineering and robotics design company founded in 1992" Boston Acoustics was founded in 1979. Th A40s on my desk came out in 1986, bu6t I'm not sure how old this specific pair are, I bought them 2nd hand. I have a pair of their bigger brothers, the A60s downstairs that I bought new in 1988...
I was about 25 when I tried to trademark General Magic and was pissed to find out I was not the first. General Products was also taken; guess there are a lot of Larry Niven fans out there.
In this case, "Dynamic" is likely referring to the fact that the robot is dynamic, not the company Boston Dynamics. In fact, Dynamics is literally just the study of forces and their effect on motion, so I feel like it would be a stretch for them to try and own a term used in every undergraduate mechanics class [1].
I figured that, but if their marketing references Boston Dynamics, like the headline of the submission did before dang changed it, I can see an argument that says that a layman might assume this project and Boston Dynamics are related in some way.
As a robotics platform, there's nothing inherently weaponizable about quadrupedal locomotion over, say, a quadcopter. Weaponizing this system would be almost as difficult as developing and then weaponizing the system.
State actors already have the weapons, so it is not complicated to attach those to a tool capable of ferrying them over rugged terrain, with long battery life, and into hiding places of human opponends. Just as they attached them to drones, cars, and other gear. What guarantees that the biggest use cases and the most motivated users of this technology are not deadly? Maybe license?
State actors aren't the threat to worry about here. They have much more controlled and reliable ways of killing you and if they decide to weaponize a mobile platform like this they are likely to fund or steal much more purpose-built technology.
If you're worried about weaponization of open source projects, your threat actors are just about any person with a computer, a couple thousand dollars and a grudge.
Boston Dynamics robot with gun:
Quote: "At the Association of the US Army’s annual conference last week, Ghost Robotics unveiled its Vision 60 quadrupedal robot, which wielded a custom-made gun atop its already eerie figure."
Yeah but non state actors don't. Just imagine dropping of robots with a gun from a car or send them via package. A 360 camera and a gun is a scary combination and as long as you use them in a suicide manner it doesn't even need to recognize friendly combatants or spare children. Just shoot everything that's bigger than rat and moves until the magazine is empty while moving patrolling the streets.
I am skeptical that licensing could stop someone from using a project as the basis for a weapon - I doubt lawsuits would seem like much of a deterrent.
That said, I am not sure that a quadruped robot is much better for delivering a bomb than an RC car, other than a slight edge on rough terrain. Both would have an advantage over flying, since weight is a problem for drones.
Eh, robots are tools, and like most, they'll be used and abused. ANYmal and Boston Dynamics Spot are currently being used for automated site inspections, and most of the top spots for the recent DARPA subterranean challenge were legged robots [1]. Of course, you also have companies like Ghost Robotics which are weaponising their platform [2]. Should we have not made airplanes because someone would eventually put bombs on them?
I'm sure there's a lot more philosophical depth to the topic, and perhaps my quick dismissal of your criticism is equally unproductive as your comment, but I see a use-case for these (frankly super cool) robots in every day society/workplaces.
What makes this work is the motors.[1] It's using 3-phase drone prop motors with no gear reduction. With direct drive, the mechanics of the joints are simple. Not too expensive, either. Each motor is about US$120.
I used to work on legged running back in the 1990s. But there were no motors good enough to build a small machine like this back then. All we had were R/C servos, gearboxes, and screw drive actuators. So I was stuck in simulation. All the early legged robots were mechanical nightmares. Springs, cables, pulleys, worm gears, harmonic drives... Now, at last, enough torque in a small package.
What you want in a leg muscle is well known - something that behaves like a spring/damper combo for which you can set the spring constant, the damping constant, and the spring's zero point. People have been trying to do that by mechanical means for decades. There were pneumatic systems, hydraulic systems, things with springs, pulleys, and wires, and a kludge called a "series elastic actuator", which is a stiff spring-damper combo in series with a screw drive actuator. They're all terrible.
"You cannot strip the teeth of a magnetic field" - GE electric locomotive sales rep, early 20th century.[2] That's why they can drop this thing and have it spring back. Try that with any of the mechanical nightmares and the results will not be good.
[1] https://store.tmotor.com/goods.php?id=774
[2] https://news.ycombinator.com/item?id=24288264