I've been following Boston Dynamics for a long time, and this story is the first one I've read which explores my primary question about the company.
I have a hard time seeing Spot Mini - or any legged robot - being useful for surveillance, for the simple reason that a thousand fixed wireless cameras will always be cheaper and will almost always give you better coverage than a single mobile robot. And I have a hard time seeing legged robots used well for moving goods, outside some niche settings such as search and rescue. Tracks are so much simpler, so much cheaper, and so much more robust. Not quite as agile, but really — how often do you need something moved over ground that a tracked vehicle or a four-wheeler can't negotiate? And that you're willing to pay $50,000 for?
Don't get me wrong, I'm a huge fan of the tech BD develops. I just don't see a business plan for it.
The National Park service comes to mind. Area's too big and disconnected for cameras + power + network (although you could sat-link the cameras, it'd be expensive).
Definitely taking care of the elderly is a huge problem we don't have a good solution for, but the technology is a long ways from being able to do that safely.
yea, also something that can convert plant into energy like animals would do would enable a robot to stay forever in such an environment without needing to recharge.
SpotMini would be great for people with disabilities. I imagine there's a huge market in healthcare for these robots. A seeing-eye or service dog cost $59,600. So if they can get SpotMini down to that price point, I think it could dominate the market.
I went to a talk by Boston Dynamics at MIT recently. They've clearly thought long and hard about the commercialization problem, and they pinpointed this exact application as the first and best domain to pursue commercially, particularly for the elderly.
I'm surprised, or perhaps, can you be more specific about what they had in mind?
If we could replace seeing-eye dogs with computer vision, good enough to help blind people cross the road, then why build it into a fake dog, not into fake sunglasses?
If you combine poor vision with also poor motor function then a seeing-eye robot that can open doors is definitely your best option.
Boston Dynamics could easily make different models for people of varying abilities. A SpotMini for the blind, one for people with motor impairments, one for people with both.
They could make one for in-home use specifically, so maybe two SpotMini's could handoff once the person leaves the house. A set of SpotMinis.
I see plenty of usage in automation. Most domestic (US) manufacturing is highly automated, there's simply no way around it. However, for some parts of the factory, tracks/wheels just don't cut it -- think iron works, rougher terrain, variable terrain etc. A bipedal or quadrupedal robot could make lots of sense.
There's also military usage -- if you can build an autonomous robot, it's not not much of a stretch to use those same design chops to augment a human.
Granted, the last one is fairly speculative, but their market is composed entirely of B2B/G and those customers are likely going to get their own highly customized versions of the robots Boston Dynamics produces for a princely sum.
Military, fire fighting + rescue, personal assistants for elderly etc (and perhaps several more). I think military applications are huge. Drones aren't very efficient and probably wont be unless they are radically redesigned.
If you build one that can carry a person, then I could see a large market for people who want to go places that they can't by powered scooter. Like anywhere that has just stairs(friends house), nature trails, etc. Buying one could be almost as cheap as putting in one of those rail elevators in your house.
Some combination of a large x8 design octocopter (4 arms, 8 motors) capable of lifting 10-20kg and transporting it up to 10km away, where it can drop a legged walking robot, could be useful for a great deal of purposes in search and rescue.
One of the things that will be a game changer for SAR robots is high-bandwidth video/remote control literally anywhere on the planet with a satellite signal, at least 400-500 kbps data rates which will be possible with the Iridium NEXT satellite constellation (vs 2.4 kbps right now). Right now such data rates are only possible in areas with decent HSPA+/LTE coverage, which is not where a lot of SAR activity takes place.
Now imagine putting a setup with a high-res visual spectrum camera, a starlight camera, and a FLIR camera on the spot mini for heat sensing/night vision, and having multiple spot mini remote controlled by people walking through trails and searching an area.
the reason legged locomotion dominates terrestrial creatures is simple: uneven terrain. otherwise animals would have wheels, since it's more efficient on relatively flat surfaces.
this is really the only (broad) use case for legged robots, which is why the military is interested in it but BD struggles to find commercial success.
I think infantry is the only answer, really. The army still finds it really essential to have lots of guys walking around on two legs, who can go anywhere their opponents can go. And often carrying a serious weight, because having stuff with you is very useful.
They have lots of tracked things and flying things too, obviously. All of which will be ever more automated, too.
I can see other niches, like carrying wounded hikers out of canyons where you can't fly, but agree that hardly sounds like a billion-dollar industry. Industrial uses I'm more dubious, I mean any warehouse already has a nice flat concrete floor... lots of uses for vision etc. but not so much for walking.
Assuming the price will come down I think about package delivery implications.
How do I get my Amazon or Uber Eats deliveries in 10 years time? Do I still go to the front door of my building, or does it arrive at my front door or even right into my loungeroom?
Or will we forever be stuck schlepping up and down from our apartments? ;)
They specialize in making robots that can react and behave in dynamic, real world environemnts. There is use cases for them all over - mostly replacing humans in high risk scenarios such as fire fighting or nuclear waste removal/cleanup etc... the tech isnt quite there yet but if theyre allowed continued funding and the time to research they will eventually get there.
I used to work with an astronomical observatory which was located on a mountain (Mauna Kea). When things were running smoothly, the telescope could be controlled entirely remotely, and the observatory itself would be unoccupied. But often some little issue would arise that would require a button to be pushed, or would require some small display to be examined and a small manual adjustment made. Now by far most of the equipment could be reset or power-cycled or programmed remotely - but not everything. So occasionally someone would have to drive up and down a mountain just to push one or more buttons. If there had been a robot that we could control remotely and have it examine a display and push a few buttons, that would have been extremely helpful. If the robot had some onboard test equipment, like a spectrum analyzer, which it could plug into a test point and produce a plot for remote display, it could significantly reduce the grunt-work of technicians. I suspect there are a lot of remote sites for microwave towers etc which could really use a robot dog that could push buttons.
What makes those buttons/displays different from the ones that can be operated/looked at remotely? Wouldn’t it be easier to just make them remotely operable/viewable, too?
Rather than improving that process adding a complex machine will only add another point of failure, which won't be noticed until the thing it's there to be used for has an issue...
I guess you could have a robot to look after the robot but that just complicates it even more
You need to look at this as solving puzzle piece by piece. In essence they are getting better and better at doing sophisticated controllers for multi-DOF robots. Atlas is 35-DOF robot and we get amazed by it. Human body is 230+ DOF mechanism. If they can continue on their current path they might be able to produce very usable robot that can replace human in laborious work. Sure, it’s long way to go but so someone would need to develop this tech nonetheless. Human labor is the largest industry on the planet worth 10s of trillions of dollars and is the biggest bottleneck. A 10 or 20 years investment for a trillion dollar business opportunity definitely makes sense.
Doesn't matter. They'll get enough DoD contracts and other sources of funding to keep them alive because certain parties (rightfully) have a strong interest in domestically retaining this sort of specialty engineering capability.
"When Google acquired Boston Dynamics, it ended the company’s military contracts. Now that the company is owned by SoftBank, Mr. Raibert said, it could also return to military work."
This stuff is not as mysterious as people seem to think. If BD robots were showing up in the military, they would be used in exercises. There would be soldiers and Marines who were training to use them. They wouldn't be invisible. This tech isn't like the stealth bomber, where the Air Force only needed a handful and could keep them out of the limelight. To be useful to the military, robots need to be deployed in war zones in significant numbers.
Fair point, I bet there is extensive testing going on at this point though. There could be considerable competitive advantage to surprising an enemy with a new technology, even though the tech appears very immature today
> extensive testing going on at this point though.
And that's where the money will come from. A dozen units here and there pays the bills when you're making super high margin stuff in barely higher than prototype volume.
They tried. Boston Dynamics developed the Legged Squad Support System, a bigger, miltarized Big Dog, for DARPA and the USMC. It worked, and was field-tested at some Marine bases. There's video of this.[1] Troops said it was too noisy and not worth the hassle. DARPA dropped the project. There was supposed to be a subcontractor to build them a custom Diesel power pack, but that apparently didn't work out, and they had to go with the noisy gas engine version.
It took about $120M to get to that point. Boston Dynamics was entirely a DoD contractor from the beginning to the Google buyout, and DARPA was very patient in funding them.
I used to follow this stuff closely. Back in the 1990s I figured out how to do slip control for legged running across rough terrain, which requires something like "ABS for legs" to control slip. I have a patent on that, and that was used by the McGill legged robot in Canada. I was thinking high-end toys, like Aibo, or the humanoid Aibo that never came out as a product. But that market segment went nowhere. I could never see a business model for this.
I met the designer of the Furby and found out what costs were like for that. No way were we going to get legged running down to toy pricing.
Today, you could probably build a good running robot toy today for the $2K Sony is charging for their new Aibo, if you could sell a few hundred thousand. The compute power required isn't that high. Big Dog runs on a Pentium-class machine running QNX, with the balance servo loop running at 100Hz and the hydraulic valve control loops running at 1KHz. Batteries and motors are way ahead of where they were in the 1990s. Boston Dynamics' new smaller robots finally got rid of the hydraulics. All this is much easier to do than it used to be.
When Google bought all those robotics companies, I thought big things were going to happen. Then we didn't hear anything for years. People in robotics thought they must have a big secret project underway. Turned out that Google had no clue what to do with all that robotics expertise. Maybe Softbank will come up with something.
Yes. A very high level of funding usually isn't sustainable unless you have some way of making money (i.e. you're a business) or you are government-funded (e.g. NASA).
The point of Bell Labs was to save the Bell System money. That's why they developed the transistor, Unix, and all the rest: because the cost to run Bell Labs was less than the cost-savings of the inventions it produced when amortized over their whole, enormous, system. I don't see any major cost-savings that Google could have gained from hella cool four-legged robots but maybe I just don't have the imagination.
This might be a bit of a stretch, but powered by the right legal team, they could very likely acquire a ton of very valuable patents that become relevant as the technology reaches a more mature state.
These robots aren't just innovations in bi/quad-pedal technology, they are making major advancements in gyroscopic monitoring, lidar, and many other areas.
Maybe not the demo robots in and of themselves, but you can bet your ass the skills and understanding they have is a very profitable business.
I would consider them elite in this field. Making robots to task based specs should pay well, and the annuity comes nearly free. Service, consulting, upgrades, changes...
Boston Dynamics mainly did DARPA contracts for a long time. And their stuff had a certain reputation... "If the contract said MTBF > 10 hours, then they pretty much ran for 10.5 hours." They really have little expertise in designing for delivering customer value, or likewise designing for user experience.
I've worked in robotics for a long time, and the thing I repeatedly see is under-attention to user experience. BD robots simply lack empathy. There are a huge number of robotics engineers who hear "lacks empathy" and snort dismissively at that criticism, and then go on to crater their robotics startup and wonder why.
The reality is that it is very difficult to come up with a robotics business model where the question: "Why does a robot deliver value better than some other solution?" has a good, clear, answer. But if you can manage to clear that hurdle, then you need to bring the appropriate design sensibilities to your product because your survival depends on it.
I'd just like a robot to be my primary computing platform. Kinda like Interstellar's robots. Elder care, personal safety, construction and last mile logistics all make sense.
53 comments
[ 4.5 ms ] story [ 110 ms ] threadI have a hard time seeing Spot Mini - or any legged robot - being useful for surveillance, for the simple reason that a thousand fixed wireless cameras will always be cheaper and will almost always give you better coverage than a single mobile robot. And I have a hard time seeing legged robots used well for moving goods, outside some niche settings such as search and rescue. Tracks are so much simpler, so much cheaper, and so much more robust. Not quite as agile, but really — how often do you need something moved over ground that a tracked vehicle or a four-wheeler can't negotiate? And that you're willing to pay $50,000 for?
Don't get me wrong, I'm a huge fan of the tech BD develops. I just don't see a business plan for it.
But really, the 4-leggers are loss leaders to pay for the 2 leggers. Compatible with human physical interfaces like doors, trunks. stairs, etc.
Much easier to care for a robot than a dog.
If we could replace seeing-eye dogs with computer vision, good enough to help blind people cross the road, then why build it into a fake dog, not into fake sunglasses?
Boston Dynamics could easily make different models for people of varying abilities. A SpotMini for the blind, one for people with motor impairments, one for people with both.
They could make one for in-home use specifically, so maybe two SpotMini's could handoff once the person leaves the house. A set of SpotMinis.
It's a massive problem space.
There's also military usage -- if you can build an autonomous robot, it's not not much of a stretch to use those same design chops to augment a human.
Granted, the last one is fairly speculative, but their market is composed entirely of B2B/G and those customers are likely going to get their own highly customized versions of the robots Boston Dynamics produces for a princely sum.
One of the things that will be a game changer for SAR robots is high-bandwidth video/remote control literally anywhere on the planet with a satellite signal, at least 400-500 kbps data rates which will be possible with the Iridium NEXT satellite constellation (vs 2.4 kbps right now). Right now such data rates are only possible in areas with decent HSPA+/LTE coverage, which is not where a lot of SAR activity takes place.
Now imagine putting a setup with a high-res visual spectrum camera, a starlight camera, and a FLIR camera on the spot mini for heat sensing/night vision, and having multiple spot mini remote controlled by people walking through trails and searching an area.
this is really the only (broad) use case for legged robots, which is why the military is interested in it but BD struggles to find commercial success.
They have lots of tracked things and flying things too, obviously. All of which will be ever more automated, too.
I can see other niches, like carrying wounded hikers out of canyons where you can't fly, but agree that hardly sounds like a billion-dollar industry. Industrial uses I'm more dubious, I mean any warehouse already has a nice flat concrete floor... lots of uses for vision etc. but not so much for walking.
How do I get my Amazon or Uber Eats deliveries in 10 years time? Do I still go to the front door of my building, or does it arrive at my front door or even right into my loungeroom?
Or will we forever be stuck schlepping up and down from our apartments? ;)
I guess you could have a robot to look after the robot but that just complicates it even more
Sounds like they haven't (yet).
And that's where the money will come from. A dozen units here and there pays the bills when you're making super high margin stuff in barely higher than prototype volume.
I used to follow this stuff closely. Back in the 1990s I figured out how to do slip control for legged running across rough terrain, which requires something like "ABS for legs" to control slip. I have a patent on that, and that was used by the McGill legged robot in Canada. I was thinking high-end toys, like Aibo, or the humanoid Aibo that never came out as a product. But that market segment went nowhere. I could never see a business model for this. I met the designer of the Furby and found out what costs were like for that. No way were we going to get legged running down to toy pricing.
Today, you could probably build a good running robot toy today for the $2K Sony is charging for their new Aibo, if you could sell a few hundred thousand. The compute power required isn't that high. Big Dog runs on a Pentium-class machine running QNX, with the balance servo loop running at 100Hz and the hydraulic valve control loops running at 1KHz. Batteries and motors are way ahead of where they were in the 1990s. Boston Dynamics' new smaller robots finally got rid of the hydraulics. All this is much easier to do than it used to be.
When Google bought all those robotics companies, I thought big things were going to happen. Then we didn't hear anything for years. People in robotics thought they must have a big secret project underway. Turned out that Google had no clue what to do with all that robotics expertise. Maybe Softbank will come up with something.
[1] https://www.youtube.com/watch?v=xY42w1w0TWk
Is there something wrong with being a Research lab that has $100B backed in funding?
I mean, wasn't Bell Labs a thing?
These robots aren't just innovations in bi/quad-pedal technology, they are making major advancements in gyroscopic monitoring, lidar, and many other areas.
Ding ding ding doesn't matter.
I would consider them elite in this field. Making robots to task based specs should pay well, and the annuity comes nearly free. Service, consulting, upgrades, changes...
I've worked in robotics for a long time, and the thing I repeatedly see is under-attention to user experience. BD robots simply lack empathy. There are a huge number of robotics engineers who hear "lacks empathy" and snort dismissively at that criticism, and then go on to crater their robotics startup and wonder why.
The reality is that it is very difficult to come up with a robotics business model where the question: "Why does a robot deliver value better than some other solution?" has a good, clear, answer. But if you can manage to clear that hurdle, then you need to bring the appropriate design sensibilities to your product because your survival depends on it.