That is super awesome and even more awesome when I read that those are direct drive motors. All the compliance is done in software. This is something I must spend some time playing with. That advent of these small brushless motors with insane amounts of torque makes all sort of things possible that weren't before.
It's the direct drive motors that make it work. As someone from GE said a century ago, "You can't strip the teeth of a magnetic field". Direct drive robot arms have been built; Adept SCARA arms were direct drive. But they used to be too heavy for mobile robots.
What makes direct drive motors work are today's really good power FETs. You can do things with FETs, capacitors, and switching power supplies to get whatever voltage and current you need, at least briefly. Motors can be overloaded by huge amounts if you monitor the temperature, and you can build up a charge in supercapacitors for the next time you need to hit the motor really hard. Putting a computer in charge of a power supply is normal today, and may even reduce the parts count.
This, and Schaft, probably spells the end of Boston Dynamics' hydraulic monsters. That approach is just too bulky for anything below mule size.
Linear motors are a lousy business. They're almost all custom, while rotary motors are standardized. Prices seem to start around 10x the price of a comparable rotary motor. Find a market for a million identical ones and the price will come way down.
Some nice linear motors.[1] Even used on eBay, they're expensive.
Is a typical example. The same gyro/accelerometer/compass 6 or 9 dof sensors from phones that the quadcopter controllers themselves use, but re-purposed into keeping the camera stable. They're really fast.
I imagine that this stuff will be great for prototyping algorithms and figuring out good applications/requirements, but eventually (like when stuff has to be mass-produced) you'll see physical structures coming back in.
Either gearwork or biologically-inspired designs, replacing some flexible electric rules and brawn with structures that have rules and brawn "baked in".
Sort of like how your knees aren't "soft-coded" for a particular set of angles, it's an emergent property of a system shaped by a particular use.
The Issus genus of jumping insects (planthoppers)[0] have a gear mechanism connecting their rear legs. The gear synchronizes the motion of the two back legs; thereby reducing 'yaw rotation' when jumping[1]. The gear communicates motion faster than a purely neurological system.
I imagine that this stuff will be great for prototyping algorithms and figuring out good applications/requirements, but eventually (like when stuff has to be mass-produced) you'll see physical structures coming back in.
I don't know about that.
This platform can be fairly cheap in mass production. The legs are really cheap. No gears. And just two motors per leg. No extra force sensors, just the control electronics.
But if you added (say) a spring into the linkages, the rest-position could be tuned mechanically to be more energy efficient (e.g: zero power). The effect of the spring could be backed out in software.
But it's not until people have played with the use-cases, etc, that one really needs to think about efficiency in terms of battery life.
But if you added (say) a spring into the linkages, the rest-position could be tuned mechanically to be more energy efficient (e.g: zero power).
That might help with one particular gait, but I think it wouldn't give as much benefit for a system as flexible as this, which can run, hop, trot, and more.
And the spring itself would need to be modeled, and would make the position and force measurements less sensitive. That's why I don't think it would be a net win.
I'm quite amazed that using springs in robots is sort of a new thing.
I was thinking about making amateur robot with two wiper motors, in-parallel, through strings, per joint. Wiper motors seem to me to be pretty fast and strong (and cheap), and use worm gear so they could hold the springs tensed without using energy proportional to tension. Keyword: "variable stiffness joint" http://www.mdpi.com/actuators/actuators-03-00270/article_dep...
The door opening and fence climbing sequence were impressive. Did onboard video, CV and CPU result in being able to do the acrobatics trajectory to hit the door handle, or was it human operator driven?
I believe it was all onboard, not human operator driven. In the article, Kenneally describes how the robot opens the door and says this: Once it perceives contact with the door knob, it retracts the leg, moves it over a little bit, and then extends it, and that actually all happens within 50 milliseconds, so it’s incredibly fast.
That's where Boston Dynamics got their legged locomotion technology. They hired the professor at McGill, Dr. Martin Buehler, who was in charge of that project.
They use direct-drive electric motors. Those are motors that you typically see on washing machines. They have super high torque but low speed which works in their use case.
A similar concept -- variable compliance -- that's been around for a while is called series elastic actuators. With geared motors you think in terms of position and speed. With direct drive or series elastic you are instead controlling torque.
Some thoughts: Looks too cute. Only two legs away from being a giant spider. Bouncing around like that, it may not be a very useful robot. Won't be carrying my drinks for instance. Or taking non-blurry pictures of anything.
