That's very nice. They finally got a more general control model. That's progress. The first flip from that crowd was two decades ago. Now they have a way of generating and following more complex motions. "80% success", they write, so there's a ways to go yet.

I wonder what they use for actuators. What you want for running and gymnastics is a spring with adjustable spring constant, zero point, and damping. A pneumatic cylinder can do that, and can store energy briefly between motions. Humans, when running, recover about 60%-70% of energy from stride to stride. (Cheetahs, 90%. That's how they run so fast.) Hydraulics can't do that very well, although there have been some attempts with hydraulic accumulators. There have been attempts to do this mechanically, with two motors and a spring, but they're bulky. Direct drive electric motors can do it, but if you have a gear train, the inertia of the motor is multiplied by the reduction ratio and you lose the ability to absorb shock loads. Also tend to break gear teeth. ("You cannot strip the teeth of a magnetic field" - early electric locomotive selling point.) I once thought that direct drive linear motors [1] were going to be the answer, but those never got popular. (Aura, the main company in that field, had some nice technology and a big financial scandal. Their web site looks like a placeholder now. It's one of those areas where hard problem + limited demand -> no products.)

There's a hack called a 'series elastic actuator', which is a powered screw jack with a stiff spring on the end, used to fake more general springs. When absorbing energy, the motor runs fast to back off the screw jack and unload the stiff spring. This recovers no energy; it uses power when absorbing energy. It's used in research robots where battery life isn't an issue.

It took Boston Dynamics about $120 million to get to the "legged squad support system", the militarized Big Dog. No idea how much it took to get to a usable Atlas.

It's good that they're still trying. The question is how much patience and money Softbank has left.

I wish Schaft had made it. They were a University of Tokyo spinoff, bought by Google and dumped by Google. They used small liquid-cooled motors, so they could run the motors much harder than usually done. Like Tesla's motors, but far smaller.

[1] http://www.aurasystems.com/actuators.html

That thing is nightmare fuel. What kind of power in voltage/amps is it on? With fusion technology, the iron man fuel cell will pretty much actually exist. Think about how much more capable these things would be with higher energies..