Many don't realize that for the first ~decade of its existence, BD was not the robotics/hardware business, but rather developed and sold human simulation software. Their main product, DI-Guy, was a modest success, and eventually provided the underpinnings for BD's core biomechanics sim tools.
When they got back into the hardware business in the early 2000's, those projects were primarily funded by DARPA/DOD contracts, but the steady income stream provided by e.g. DI-Guy helped fill the gaps between contracts, allowing them to run a business that, though lean, stayed in the black. Eventually, as a part of the Google acquisition, DI-Guy was divested to VT MAK, another local company in the modeling/sim software business .
Source: former employee that worked at BD for the better part of a decade from the mid 00's to late 10's
As someone who is in the field of mobile robotics, what Boston Dynamics does is often way ahead of its time. Robotics is hard and expensive. If you are developing a new form of locomotion it will take years and millions of dollars before things work out. Heck, I myself have seen my colleagues spend 10+ years just ironing out issues with underwater vehicles before they even see the market.
This is a trend that gives me hope for the future. There's ever-increasingly advanced technology and technical knowledge accessible/affordable to the public, both hardware and software. What used to cost the military millions of dollars to develop, are now mass-produced and distributed, to fit in a pocket or to build one yourself.
Sure, the cutting edge stuff will always be in the realm of experts with funding, but the pace of progress in technology is such that there seem to be plenty of opportunities for innovation from the grassroots level, serious amateurs and hobbyists.
It kinda feels similar to the early era of personal computers, where anybody with interest and ideas could enter the field, learn things on their own, and even contribute to its advancement.
I don’t think it’s advancing technology that’s allowing this. It’s not physically constructing or programming a robot that’s so expensive, it’s iteration of designs to get to that point. A hobbyist can build a small working jet engine at home, but their hardly starting from scratch.
Further, building a modern full scale General Electric GE90 is extremely far beyond what a hobbyist can accomplish. If anything the true cutting edge is getting ever further out of reach. A few peasants could build a full scale trebuchet, good luck getting your friends together to build a F-22.
That just seems to move along in little tiny steps. No balance or slip recovery shown. That's the wind-up toy level of walking.
Still, it shouldn't be that hard to make a small version of Big Dog technology now. They put a lot of money into fast hydraulic servos, which you no longer need for anything smaller than a horse. Gyros and accelerometers are cheap now, because they're phone parts. Motors and motor controllers are much better and cheaper now.
Big Dog's compute power was a Pentium 4, which is nothing today. Force feedback sensors for the feet are still expensive, because there's no mass market. You need those for slip control.
They're basically a research and development firm. How many research teams do you know that are actually "profitable"? Money goes into research because it may pay dividends in a few decades, not because it's profitable in the first couple of decades.
The markets and VC are too short-termist for things like this. The failure of many VC-funded robotics companies is evidence of this. Rethink Robotics. Jibo. Anki. I can keep going. The VCs ran out of patience long before the product gained market traction.
And Boston Dynamics is not an exception. Government funded primarily until 2013 when it was acquired by Google. Google then didn't have the stomach to continue to fund it, so they sold it to Softbank.
IMHO, what BD really needs is a specific target use case to build towards. Right now Spot/Handle/Atlas are being positioned as robotics platforms. They're leaving it entirely up to others to figure out the use-cases based on the platform.
However, without a really salient use-case to demonstrate why a $75k quadruped is better than a similarly priced remote-controlled and tracked robot, it's hard to envision how these things will take-off. The use-cases that I've read offered "search and rescue, bomb disposal" remind me of how early personal computers kept advertising "recipes and electronic check balancing" before they found their potential.
What if, and here's a stupid example, BD knuckled down and super optimized Spot for truck-to-door package delivery. What if the driver never had to get out of the truck, stop the engine, or secure the vehicle to go walk the 30 feet down my driveway to drop off the last package I ordered? What if Spot simply grabbed the package, jumped out of the truck, dropped it on my doorstep and the ran back to the truck and repeated for all the houses on my street? What if over a period of some number of years it could be shown that this increased efficiency for a truck to more than cover the cost of the $75k robot?
Spot would have to be able to:
- charge on the truck while in transit
- autonomously identify packages it can deliver (reading barcodes)
- have a GPS database of all front stoops then navigate safely to them with a package
- find it's way back to and onto the truck
- alert the driver when a package is too large/awkward for Spot to handle by itself,
- know when the driver decided to deliver the package instead of Spot
You lose some truck volume to Spot, some batteries and a charger. But as the truck fleet electrifies, it could just use the truck batteries for some of this.
Let's say each truck upwards of 200 stops on a route , at 1 minute per stop. A Spot enabled truck could drop that down to 30 seconds per stop. At the upper end (200 stops per route) that saves more than 1 hour of delivery time. Assume we're not trying to increase the carrying capacity of the truck, but reduce the hours worked per driver. UPS drivers make around $32/hr . With Monday - Saturday delivery, that's about $10k per year per driver saved. At current Spot prices, that's about 8 years of operational service to pay for a Spot. Assume at volume the price per Spot can drop down to $50k per robot. Then that's a 5 year return.
Let's say that this system works for $35% of all UPS and Fedex routes (delivery to single homes and townhouses) and there's one truck per route.
Fedex - ~30,000 total trucks = 10,500 Spot enabled trucks
Total of 45,500 Spots = market of $3.4billion for BD at $75k per Spot, $2.27b at $50k. BD could probably do somewhere in the middle if they just leased the fleet of Spots, and provided managed maintenance, upgrades, and other services. What if USPS could take part in this, they operate ~140,000 LLV (mail trucks)?
Now let's say these fleet delivery operators get self driving delivery trucks. How much of all package delivery could be handled by an automated truck + spot system and how much does that save UPS/Fedex in salary? There's no point making the trucks self-driving if the last leg of delivery isn't solved since they then have to pay a premium for the trucks and still pay for the delivery human, so a Spot-like solution would have to happen first.
I have to imagine that blanketing an area with seeds dropped from a flying drone would be far more effective than a ground based robot which planted trees. Obviously the success rate of a ground based robot would be far higher, but I think we'd be better off going for quantity over quality.
If that assumption is true, then we already have the tech to do this.
I recall when HN was discussing dropping seeds from an AC-130 few years ago, someone pointed out that it would be much more cost-effective to just hire locals to do it by hand. The same will probably apply to drone-planters.
Seriously, I wonder how the engineers there will feel when these are sold to the military or the police and weaponized. Maybe they say they won't now, but to say it will never happen is wishful thinking IMO.
Likely as guilty as we will when our disposable gadget, data center power hungry environmental mess is left to the next generation.
Hard to really get behind such customized moralizing when we’re not really putting ourselves aside as engineers doing grave damage but on a longer, vaguer timeline.
Our linear progression towards yearly updates is hardly warranted anymore. Exponential breakthroughs are pushed to the side with our fetishizing stable revenue over R&D, low taxes, unicorns built on cut/paste-able software IP, and mathematical gamesmanship, and eyeball addiction over variety of ideas.
Purposefully stimulated chemical addiction to agency of self indulgence and nihilism is generating economic activity, not novel invention.
Take some solace that it’s not all your fault: We’re not allowed to assess the validity of many key economic assumptions, even with historical evidence like the pre-80s era of high taxes and strong unions.
Just trust that non-gods picked correctly for all time decades ago, and iterate on their meaning!
Really having a hard time with the glaring hypocrisy of all first worlders right now. Science! Oh but also yeah just leave economics alone, the physics there has to be read different.
Take yourself a bit more seriously first and we’ll moralize about others later.
Would be cool to have a mobility device based on their handle robot (the one with wheels). Basically a hybrid of bicycle and wheelchair that can go over stairs, and be safer in case of accidents on the road.
> Would be cool to have a mobility device based on their handle robot (the one with wheels). Basically a hybrid of bicycle and wheelchair that can go over stairs, and be safer in case of accidents on the road.
Didn't Segway demonstrate something like that?
Edit: Okay, it's only somewhat Atlas-like, for very low values of Atlas. Dean Kamen first demonstrated the iBOT wheelchair in 1999:
Backwards compatibility is one of the reasons, meaning an humanoid form can interact with all the interfaces that humans already use in day to day basis, think push/pull doors, grabbing two-handed objects or pushing a car; as well as the ability to be aid by other machines without wasting their own energy: think using electric stairs, elevators, boarding a car/bus/train, sitting down at a computer desk (so hands stay on while legs are off). Plus doing all that without the risks that spinning blades bring.
Because then suddenly it's a tragical accident caused by a software fault, with no human responsible - just as most data breaches are framed as "cyber attacks" instead of what they actually are: negligence.
Horses need to be trained (usually they don't like trampling people) and have high maintenance cost. As for the wheeled vehicle, it can be prevented access with roadblocks. Also it can't enter houses. Something that can walk over piles of rubble would be clearly very useful.
Cops manage to permanently injure or kill people with rubber bullets. I'm not really sure I can agree that the malice is in the tool and its creator. No matter how benign the tool you create, someone's going to abuse it.
Should fishing companies feel responsible because someone out there used fishing line to strangle people to death?
Dang unpopular opinion on a tech forum I guess but I'm with you, and if I show the Boston Dynamics stuff to anyone not a tech enthusiast I'm likely to get a "why are they building executioner robots"
It's like Elon's global internet. There's definitely some lives to be improved with video chat deployed anywhere on earth, but I don't take that to be a benefit worth the cost of enabling worldwide real time surveillance and the deployment of remote control killer robots.