Hacker News Reader

Comment by danvoell | Hacker News Reader

Pure speculation here. I can't imagine that the "toothbrush test" is accurate here. There is a large market for robotic arms. And if Google's robot creates any advantage over existing robots in manufacturing, Google could go up or down stream in the verticle.

My guess is that it hasn't been tested enough and/or hasn't cleared the top level of safety in the market. One incorrect swing of a robotic arm could severely injure someone and that is a huge liability for Google's name.

View on HN

"There is a large market for robotic arms."

There is a small market for robotic arms. Total sales in 2014 were 229,000 units worldwide. Worse, from a Google perspective, robot arms require a substantial support infrastructure, with on-site installation, maintenance, and support. Google is terrible at support, let alone the kind of support where the customer wants a service tech on site right now because their production line is down. That's the norm in the industrial robot business.

Rod Brooks' "Baxter" robot was supposed to change this. Introduced in 2012 with a lower price than competitors, enough AI to make it easier to "teach", and better safety around humans, it was going to be the industrial robot that sold like a mass market product.

In 2015, only 400 "Baxter" units were sold.[1] The company's new "Sawyer" robot is said to be selling better, but hard sales numbers don't seem to be available. This just isn't a big industry. The first industrial robot, the Unimate, was sold in 1961, and there have been wildly optimistic projections for over half a century now. But the actual numbers aren't impressive.

Google acquired all those robotics companies and accomplished nothing with them. Even the ones that were shipping a product, such as Bot and Dolly, stopped shipping.

[1] https://robotenomics.com/2016/01/11/the-facts-about-co-bot-r...

What went wrong with Baxter? I remember it being so revolutionary when it came out.

Baxter is supposed to be safe around humans, but to achieve this, they had to make it weak and slow. So it underperforms competing industrial robots. Compare: Baxter and Sawyer: [1] Festo: [2].

It's also not precise enough for precision assembly. Baxter videos show it stacking medium sized parts, not assembling cell phones.

[1] https://www.youtube.com/watch?v=vPth0Q4apLU [2] https://www.youtube.com/watch?v=aH_t_1-tl40

In between you have stuff like Universal Robots' stuff which is human safe and relatively fast. For a startup knowing that the arm you're using isn't going to kill anyone[1] is huge in terms of being able to develop stuff quickly.

https://www.universal-robots.com/

[1] Or kill yourself with it, I've managed to hit myself in the head but didn't even get a bruise.

I have a Baxter Research Robot. The theory was that you would make up for the imprecise gear train (backlash at the gripper of 10-20 mm) with visual servoing. But this turns out to be hard. A big limitation is that the gripper-mounted cameras are 30 FPS (less if you try to use both) and add a full frame of delay, so the visual servo loop has to be really slow (like, 1/2 second to zero in on an object). That means it's far less productive at simple operations like loading parts into boxes than humans.

Also, its industrial design is really bad. I hate its hulking arms and smiling animated face. It's surprising that you can ruin an industrial product with ugly visual design, but there you are. For another example, see the Boeing X-32. https://en.wikipedia.org/wiki/Boeing_X-32#/media/File:USAF_X...

Rant: Sheesh, now even Wikipedia is using full-page JS popups that fail to load and make the page unusable. Seriously webdevs of the world, do we actually test our own work on phones?

At least I was able to see the image first. I was wondering how much of the X-32's shape was constrained by functional requirements, and the same for the Baxter. Can a safer arm be made without looking as you describe?

The X-32 and X-35 were two candidate designs for the Joint Strike Fighter. The X-35 was chosen. They had the same functional requirements, but the X-35 looked less silly.

https://en.wikipedia.org/wiki/Lockheed_Martin_X-35

You could easily avoid having a decorative face. The general arm configuration might look OK with some other ID.

What do you use yours for?

Machine learning research. But I'm currently using Fetch and simulations instead.

I own a Panasonic Robot welding cell, and I keep up with numbers of robots sold like that - I know your numbers are correct. My experience with programming the Panasonic is that there is quite a learning curve to get over, to do the most basic things with it. I think the small numbers of robot adoption could be a reflection on this difficulty, combined with high cost. My cell was $85,000. The robot was a great value at that price though, it generates about $2000 an hour worth of product. Paid for in 40 hours of welding, good for 10,000 more hours before its worn out.

The Baxter should prove me wrong, but I think it is just an underperforming platform. The current owners of robot tech would view it as under capable compared to arms from Kuka or Motoman or Panasonic etc. www.youtube.com/watch?v=tIIJME8-au8

I think the non-robot users, those companies getting a first robot, would find Baxter easy to use but would be less inclined to go buy one; getting a robot is a big step.

I have to think that ROS getting better and better, with offline programming, will help adoption. We are at the UNIX terminal phase of computing, or at best we are at a Windows 3.1 level. Nobody has made the iphone of robots yet, one that is very capable yet easy to program. Once we can get to a few swipes to get result, robot placement will increase rapidly. I love how repeatable the result is, and how fast it produces parts. It is about 3x more productive than my most productive welder, with equal or better accuracy (we still hand weld a lot of things)

Baxter was supposed to be the super easy to program robot, but he never seemed very precise in the videos, nor very fast (by design, but still - we buy these things to make money).

Kuka, the world's largest robot arm manufacturer, got acquired for $5B [1]. I'm not sure how that's a "small market." If Google believed in robotics (beyond autonomous cars), which seems to be the case given the acquisitions, then they need to start somewhere.

Universal Robotics was demonstrably worth $300M with (what sounds like) a more-or-less similar product. That seems like a good place to start -- and all big unicorn products start somewhere; very few jump straight from $0 to $Billions.

[1] http://www.wsj.com/articles/chinas-midea-group-announces-bid...

Google's market cap $500+B so if they hit a home run their market cap will only go up 1-2%.

Until Google created AdWords, you could say the same about online search and sem ads too. All frontier markets start small and become big over time.

One of the biggest robot companies in the world, and its only worth $5 billion? Yes, it is a small market if that statement is true. Which is a shame. Kuka robots are amazing, and capable of welding together cars to being rides at Legoland. And yet, one of the premier robotics companies in the world is worth less than 10% of Uber.

Robots are still hard to use, hopefully soon this will change. Then the market will get big. The robot arms are already precise enough, they need volume manufacturing to get cheaper.

When we get the iPhone of robots, we're going to see a huge societal change.

I've been waiting for the IBM PC or Apple II of robots for far too long, and it's still not here. It's frustrating. Robots just are not very general purpose yet.

From the article, the second half of the toothbrush test is 'used daily by billions of people'. Can you show the robotic arm market that large?

With self driving cars, I believe Google is showing willingness to develop things that could severely injure someone.

disclosure: I work for Google. I have not been involved in toothbrush test discussions and do not work on robotics nor cars.

Yeah but what if there's one machine that makes billions of toothbrushes. Does google decide to not make that machine?

Potentially, sure - as the article mentions at present, AI isn't able to be involved in moving the arm at all. How long until it isn't? What if you could have an arm and sensors installed in your kitchen that would cook dinner for you, or wash dishes? Sounds complicated...but more than a self driving car? There's a lot of people that would (and do) pay a lot of money to have someone else prepare and clean up their meals.

Moley is trying to make a robot chef and release it in 2018 [0]. Personally, I think it will bomb because the AI isn't there yet. When a grip on a tool or bowl fails the whole thing fails. Robotic arms aren't "set and forget" when dealing with even the simplest failure modes. That said -- successful robots like this are why the potential market for robot arms is huge. The current market for robot arms is very small. The gap between the two is an AI problem.

[0]https://www.youtube.com/watch?v=QDprrrEdomM

Sure. In the future (~20 years) I expect that a robot around the price of a luxury car will be able to perform tasks such as laundry, cooking, and cleaning (not just vacuuming but nearly all types of cleaning). These robots will be used in workplaces and homes, and the eventual market size is definitely in the billions. These robots will need arms, of course. These arms will be sold by a company that releases a small-volume product early and iterates, not by a company that waits until they have a billion-person product before releasing anything.

To be fair - this 20-year-prediction-to-home-robots has been happening since the Jetsons aired in the 1960's.

If anything, the Roomba proved that there is small niche-market demand for robotic automation in the home, and economies of scale for goods follows the demand. Consumers simply aren't going to spend the equivalent of buying a new car for something that can be done by a minimum-wage maid service.

"Consumers simply aren't going to spend the equivalent of buying a new car for something that can be done by a minimum-wage maid service"

a leased new car costs the consumer a few hundred dollars a month, which is the equivalent of say 30 or 40 hours of minimum wage labor. a robot that cleans floors, prepares food, washes clothes, folds laundry, cleans windows, serves meals, puts dishes away, feeds the dog, waters plants etc. could work 24/7 (with swappable batteries). i'm not convinced that consumers wouldn't pay that if the technology could get there.

of course we are far from a robot that can do all of that, but in my opinion, the prize to the company that can deliver this is massive.

Consumers don't lease appliances, and they're certainly not going to pay $500/mo + maintenance for something that they and/or a maid can do.

It's an inferior service, the same reason the Roomba didn't put Dyson or Hoover out of business.

They're better off trying to sell a "real" phone and "table-bot" as their first venture into hardware.

Also, how big would the market be if you could just say pick up that widget and put it over there on that belt. And, it said no problem boss.

And by the way it worked with any slim margin robot arm.

Robots will be stocking shelves before they're in homes. And they'll be picking off of warehouse shelves before they're stocking. They're already moving stuff around in warehouses, like what Kiva is doing.

The comparison in the article was to Universal Robotics, whose arms are low-payload and compliant (ie. not big 'ole position controlled arms of yesteryear).

Toothbrush test generally refers to daily use. But I'm not sure I understand that context outside of mass-market consumers. A well-designed arm for light-manufacturing has 90%+ uptime and is (in fact) used daily.

Reply on news.ycombinator.com