You're comparing the GE quad with the BD quads. The video you link to mentions that the mental load of constantly thinking about four feet means the operator of the GE quad has to stop after about fifteen minutes.
Now the BD quads can run around untethered, with a human operator who is presumably doing little more than selecting gaits on a remote and steering it. If I make the least charitable reading of the Cheetah video at hand, it's still a damn long way from being useful, but it's closer to being useful than that giant truck-beast - imagine this a few iterations down the line, married with the results of self-driving cars mature enough to go offroad, for instance.
There are people working on all of the things you mention. And there are people playing with what can be done now. Your attitude seems to be "if I can't work on near-human robots, I'm not interested", which is fine, but do you really need to shit all over everyone else who's decided to see how far they can get with what they've got right now? It took evolution about 1800 million years to get from single-celled life to humans. We've only been trying to build human-mimicking automatons for a few hundred years. Have some patience. Stuff like the folding cubes robot? That's the pre-Cambrian explosion of wild forms. Which ones are practical? We won't know until we've tried 'em.
We'll get there eventually. Maybe not in my lifetime or yours, but we'll get there.
Perhaps I failed to communicate my position clearly enough. Speaking of something like Wildcat specifically, I don't see the point in repeated mechanical engineering exercises that are expensive and do little to move the art forward. To me that is a waste of money. Take the GE platform and evolve it. Get rid of the human controls and integrate a modern compute engine. Add sensors. Now you have a platform for significant work in areas that need it. Show me this platform navigating a Home Depot while attending to the various products they sell. Show me the platform searching a pile of rubble for survivors. Examples abound. Few, if any, require expensive mechanical design exercises to reinvent the wheel. In fact, you could probably do a ton of good development with a good size quadruped constructed out of large RC type servos and linkages. All of the computing would have to be external, but that's OK. Build one of those that can do what one of my dogs does (use smell to find objects) in a suitably complex environment.
My problem is with the idea of throwing tons of money in the wrong direction. We don't need to design quadruped after quadruped that perform narrow parlor tricks. We need better brains, sensors, actuators, materials and a myriad of other area improvements that will make a true difference.
Isn't that… kinda what BD is doing? I'm sure they could give you a list of prior work they're building on as long as your arm. Their main current work seems to me to be in building self-stabilizing walking platforms, upon which they can eventually carry various other loads, such as, oh, the dog-level AI/sensor package you're using as an example.
I dunno, I guess I kinda want to take on the role of the Voice of Startup Culture here and say "congratulations, you've identified a hole in the market, start working on your MVP". Except I also kinda feel like that's exactly what all the modern work you disparage is doing; the MVP of "robots" is a huge, multifaceted problem, and nobody's made it out of the "cool tech demo" level yet.
You've made a list of a half-dozen fields that need to be advanced. Which one are you working in? Either as your day job, or your hobby? Or are you just sitting on the sidelines going "it's all been done before"?
> You've made a list of a half-dozen fields that need to be advanced. Which one are you working in? Either as your day job, or your hobby? Or are you just sitting on the sidelines going "it's all been done before"?
Let's be careful not to engage in shooting the messenger, which does nothing towards addressing the validity of the argument.
That said, I am playing in a couple of areas, for example, applying AI to learning. The huge difference is that I am doing it in the context of a private enterprise. I am not asking taxpayers to shovel money at my toy projects. There are companies out there that exist solely because of hundreds of millions of dollars of tax money being thrown at projects that produce very little of real value.
What would you say if DARPA threw tens or hundreds of millions of dollars at a company that set out to rebuild an operating system virtually identical to Linux? I would hope you'd come forward and say that we need to work on other things. This has been done, multiple times, and does not need to be reinvented until advances in other technologies warrant it.
If you read what I said carefully you should see that my primary argument is that we are throwing money at the wrong problems. Tax money.
If you have a private company and want to iterate through a dozen expensive mechanical design exercises for whatever reason and YOU fund it. Fine. Your money. Your decision. You get to do whatever you want.
My problem is with tax money that is repeatedly thrown at the wrong problems. We don't need to push for advances in mechanical design. We need to push for advances in the areas I mentioned and more. Sensors, for example.
It's open source. You can download the entire design and build one yourself. I have it on my machine. I can open every single part in Solidworks, look at it and modify it.
Tell me what this platform is lacking and why we have to spend so much money reinventing the wheel? Hundreds of millions of dollars.
We and others have already thrown lots of time and money into the mechanical design of a myriad of platforms that are fantastic for the development of the really tough areas that need attention. We somehow choose not to build on top of that but rather throw money at doing the same thing over and over again and producing little usable tech out of it.
Do you realize that Boston Dynamics got nearly TWO HUNDRED MILLION DOLLARS of US government money [0]? For what? To build otherwise useless platforms that make the government guys go "ooh" and "ahh" because they really don't know what they are looking at. It's magic! Let's throw more money at it. While, in the meantime, the money should have gone into some of the areas I highlighted and others.
This is not the best analogy, but I'll try. This would be similar to throwing $200 million dollars to repeatedly build MRI machines that do little towards addressing the needs of, say, Cancer research. In reality you need to throw the money at specific areas of Cancer research (not a biologist, so I can't list them) rather than at making MRI machines. We know how to make MRI machines. That's a done deal. What we need to improve is everything else.
We know how to build robots like these. These are academic exercises. Take a small team of engineers covering the required fields and they'll build you these robots. It's an exercise in IMPLEMENTATION. We know how to do it. It's just a matter of doing the work.
How many HN members could build a little galloping robot in, say, five years, if I threw five or ten million dollars at them and there was not requirement to build something that was commercially viable in any way. Lots. Same results as BD with 5% --or less-- of the money BD got. That's a sign that this requires no more R&D money until we improve everything else.
What we need to invest money on are the things we don't know how to do well enough. They are not very sexy. I get it. Who wants to be the guy to throw money at AI for ten years and have to explain to a General why all you got are some graphs and numbers on a screen. Significant as the results might be, it's a lot easier and more exiting to show a bunch of ignorant gate-keepers a ridiculous machine galloping along or bouncing around. It's a sexy supermodel in a bikini. Very soon everyone forgot what the hell they were there for and throw more money at it so they can see another one.
Now the BD quads can run around untethered, with a human operator who is presumably doing little more than selecting gaits on a remote and steering it. If I make the least charitable reading of the Cheetah video at hand, it's still a damn long way from being useful, but it's closer to being useful than that giant truck-beast - imagine this a few iterations down the line, married with the results of self-driving cars mature enough to go offroad, for instance.
There are people working on all of the things you mention. And there are people playing with what can be done now. Your attitude seems to be "if I can't work on near-human robots, I'm not interested", which is fine, but do you really need to shit all over everyone else who's decided to see how far they can get with what they've got right now? It took evolution about 1800 million years to get from single-celled life to humans. We've only been trying to build human-mimicking automatons for a few hundred years. Have some patience. Stuff like the folding cubes robot? That's the pre-Cambrian explosion of wild forms. Which ones are practical? We won't know until we've tried 'em.
We'll get there eventually. Maybe not in my lifetime or yours, but we'll get there.