as someone who did a bit of graduate research in that area, i would debate the "really, really well engineered" part of that statement. they brute-forced the engineering and developed energetically expensive machines that were great as demos but not practical as products. compare how long one of their robots last vs any legged creature of comparable size, and the poor energetics are obvious.
i would speculate that their company was a bet that battery packs would become light enough yet have a high enough energy density that it's robots could one day become viable products, but it doesn't seem like they're winning that bet.
that's in contrast to the segway, which was a bet that computing power was fast enough to dynamically adjust and balance human scale motion with high dynamic stability. they won the technical bet, but still lost the productization bet.
There isn't any particular inefficiency in their walker balancing implementation.
but if we just look around, we find a huge array of locomoting, self-contained, relatively energy efficient beings to draw inspiration from.
for example, the tendons and muscles in your leg can return as much as 40% of the energy expended in locomoting. and your legs include passive mechanisms (like how your knee joint locks and the alignment of the cruciate ligaments inside it) to keep you mostly upright without a lot of active control.
nature has ingenious solutions to learn from and take advantage of.
Irrelevant in a factory, where you can adapt things as much as you like - but crucial if you envisage people sending robots to environments they don't own - like a robot carer helping Grandma go to the store.
Needless to say, the profitable applications for this aren't especially short-term, so the ROI for this research is hard to predict.
If you prefer, you can substitute "Like a robot delivery driver, who must deliver to flats without lifts" or "Like a robot cop, chasing truant teens through a mall" [1]
They cost about $7/hr each to employ (~$62k annually since they're 24.7): https://www.digitaltrends.com/cool-tech/knightscope-robots-i...
A robotics mentor once told me "just use bigger wheels" for rough terrain rather than legs and fancy control algorithms. But stairs are particularly difficult for wheels. House bots will almost certainly need legs to maximize utility. Unless they are so cheap you can get one for every surface.
Tell that to these guys: https://www.youtube.com/watch?v=v1apqfrJYwk. I think one day we have have two wheeled robots with these kind of capability.
That's one, and also a much smaller footprint than a wheeled vehicle.
> This how modern salad farming happens for instance.
And mushrooms too.
Mushrooms too.
1. wheeled sidewalk robots. 2. wheeled robot with arms 3. walking robot.
Still incremental just have sellable robots the whole time.
The US Army was also trying some exoskeleton designs. The Raytheon XOS [2] was quite capable but needed an external power cable. The Lockheed HULC [3] finally reached a self-powered configuration, but battery life was too short.
None of these will be fielded. Maybe the next generation.
The Army's next big "mobility" buy is a replacement for the HUMMV, an armored truck from Oskosh. Smaller than the MRAP, but a comparable level of protection. Boring, but useful.
[1] http://www.military.com/daily-news/2015/12/22/marine-corps-s...
I imagine we'll see progress on this front when there's a breakthrough there. Barring that, you get some analog of the rocket equation: power source + fuel requires armor (for survivability) which means more power and fuel is required, etc etc
Another avenue to cheat: put extra armor just around the fuel/power source.
Wouldn't that make it heavier and take away the advantage of having an exoskeleton in the first place?
I imagined something like the Soviet "Shtrafbat", battalions made up of prisoners and other undesirables.
Of couse the shtrafniks will die in a terrible conflagration when they bump their battery, but think of all the expensive equipment taxpayers can buy for them, and a use for the prison population! Of course the prisoners will be rented to the state for a nominal fee, and there will be compensation to PrisonCo when their property is damaged. It's a win-win-win!
Fact is that Boston Dynamics does one thing, really, really well engineered dynamic locomotion. From what I understand there's very little ML/AI to it, just really good modeling and engineering.
However, no one has quite figured out exactly where this technology will be most useful or practical in the near future. Even (maybe especially) when they were owned by Alphabet.