So what's the main barrier to using these industrially?
Is it reliability? Did this video a zillion takes?
Is it manufacturing? Maybe it's hard to scale them up?
Is it battery life? Maybe it can power itself for just a few minutes?
Is it object manipulation or sensing/world understanding? Maybe it can't apply the right forces to a soft thing or a flexible thing or know what it can step on vs over or what will move or stay still?
Is it just that anything worth automating is worth specializing, and there are better robots for different specific tasks?
We've seen industrial robots for a while now, and these more general robots have gotten really good, so where are they?
In actual factories, which are highly controlled environments, the things called robots are much simpler and more specialized. So you're "Is it just that anything worth automating is worth specializing, and there are better robots for different specific tasks?" point is more or less it.
But also consider, while these bipedal, quadrupedal and wheeled robots can do all sorts of these in isolation, their ability to accomplish things autonomously in the chaotic, unstructured world outside the factory is little-to-none, accent on none. The Darpa Robitics challenge was essentially considered a failure, all entrant failed. Most could not do the "walk to a door and open it" challenge.
Similarly, Boston Dynamics sells their "big dog" walking robot to the military but it is seldom if ever deployed. It's strong and faster than a horse with basically the use case as a pack horse. But well trained pack horse won't injure if you get in front of it and will walk along with the troops on it's own without a guy with joy stick directing it. And the Big Dog needs constant direction.
Basically, robots don't have even the "animal intelligence" needed for real world activity.
And the Big Dog requires a lot of technical experts surrounding it at all times, an energy production system, and if it falls over or gets stuck it will be harder for one or two people to get it back to orientation.
Until there's light weight batteries or fuel cells available, big, long-endurance robots like Big Dog just aren't practical (especially for military use).
> So what's the main barrier to using these industrially?
In an industrial setting, a purpose-built robot will outdo these robots, or a human, at nearly all tasks. And whatever these things are doing, likely a human can do it better with a little bit of training- that's low capital cost compared to these robots.
Carrying things around? Conveyor belts. Picking things up? Robot arms. Going up and down stairs? Elevators and/or conveyor belts. If your aim is to make money, a general purpose robot is rarely the best choice.
The only real purpose I can see for these things is being able to travel quickly in rough terrain. Military applications, basically. Just picture a small army of these robots, armed with tasers, chasing you through the woods. While dancing.
> The only real purpose I can see for these things is being able to travel quickly in rough terrain
This reminds me of the (alleged) Tom Watson quote: "I think there is a world market for maybe five computers."
These robots will be everywhere in twenty years. The cost and the software aren't there yet, but it's only a matter of time. Flexibility and time to market will beat purpose-built robots in many cases once the cost comes down. Many purpose-built factories will disappear. Local, distributed generalist factories will spring up, staffed by robots. Drones and autonomous vehicles will deliver "hand-crafted" products directly to customers.
With autonomous vehicles poised to replace the most common job in the majority of U.S. states -- truck driver -- we're on the verge of massive systemic changes. That's at best. At worst, with economic and political tensions already high, we'll get violence and bloodshed on a global scale.
> This reminds me of the (alleged) Tom Watson quote: "I think there is a world market for maybe five computers." These robots will be everywhere in twenty years. The cost and the software aren't there yet, but it's only a matter of time.
Hitting the nail on the head here. I've been following Boston Dynamics for the past 10 years and it's just amazing the progress they've made and it's going to get augmented massively with better battery technology and ML/AI. No doubt in my mind these types of robots will be everywhere in 20 years as you say.
> These robots will be everywhere in twenty years.
Doing what? What can a generalist robot do that a task specific robot can't do better, faster, and cheaper?
I'll be happy to be wrong- they're really neat- but I spent 7 years in the Operations tech division at Amazon, working in warehouses especially, and I don't see any need for these robots vs alternative options.
> What can a generalist robot do that a task specific robot can't do better, faster, and cheaper?
Nothing. And everything.
Nothing, because no matter what task the generalist robot is assigned to, it'll always be theoretically possible to design a custom-built specialist robot that's better at that one task.
Everything, because no matter how much better your sock-drawer-organizing robot is at organizing socks, it'll never be able fold my clothes or make my bed or clean my windows like the generalist robot can. I'm not going to purchase thousands of different robots for thousands of different tasks when I can just buy one that will do all of them.
A human in India or China can be had for a couple of hundred dollars a month. These robots probably cost hundreds of thousands of dollars to buy and maintain. And they aren’t as adaptable.
I don’t think this is it. US farming is vastly more productive than in India or China. It has less to do with the cost of labor but the nature of labor, ie these robots don’t have a niche.
> Is it object manipulation or sensing/world understanding? Maybe it can't apply the right forces to a soft thing or a flexible thing or know what it can step on vs over or what will move or stay still?
Did you notice the robot's hands? Or, you know, lack thereof?
These robots are impressive, amazing and advanced ... but they are demonstrating mobility, not utility.
If you want your robot to really replace people, it needs amazing hands. Also amazing visual processing, but there's a lot of work being done there by the autonomous vehicle people, and I'm not sure who is currently leading the pack on hands.
Lots of interesting answers to your question, but if I may present another:
There is none.
These robots are currently in use screening COVID patients, assessing parts of the Chernobyl plant and working in warehouses. Those are just the applications that I know of.
I think the "barrier to industrial use" was simply that Boston Dynamics wanted to ensure they had a fairly mature platform before scaling up to significant production.
Repetitive physical tasks can usually be done with existing machinery and industrial robots. To do anything more complex than that, i.e. non-repetitive tasks, even if they could do it physically, they lack the required brain power. They don't follow verbal instructions except very basic ones.
Price. Only the price is the barrier. :)
But it is getting better quickly as we speak.
Spot is 7000 USD. This is cheap enough to be useful in many situations, but certainly not in your average storage facility.
Who needs dancing robots? What I mean this video does not show that useful work has been implemented. Probably flexibility of tasks might be a problem. Or there is no problem and we see rollout in the next few years.
It shows that they can fuse IMU and other sensor data and feed it to a trajectory-optimization system and get actuation torques meeting realtime contraints sufficiently quickly to not fall over. Bonus points for smooth motion.
Check out MIT’s Underactuated Robotics class taught by Tedrake. Spring 2018 is all on Youtube, I believe. A Raibert Hopper simulation is one of the first homework sets. Post a link to tbe video of your solution.
Is it reliability? Did this video a zillion takes?
Is it manufacturing? Maybe it's hard to scale them up?
Is it battery life? Maybe it can power itself for just a few minutes?
Is it object manipulation or sensing/world understanding? Maybe it can't apply the right forces to a soft thing or a flexible thing or know what it can step on vs over or what will move or stay still?
Is it just that anything worth automating is worth specializing, and there are better robots for different specific tasks?
We've seen industrial robots for a while now, and these more general robots have gotten really good, so where are they?