Hacker News new | past | comments | ask | show | jobs | submit login
Google Puts Money on Robots, Using the Man Behind Android (nytimes.com)
186 points by ipince on Dec 4, 2013 | hide | past | web | favorite | 81 comments

I wonder how long it will be before the sex industry catches on. Surely realistic and anatomically correct androids are preferable to blow-up dolls are what have you... even Star Trek mentioned the issue.

It might seem flippant, but I genuinely think that is a potentially huge market (though probably not Google's cup of tea). After all, why create a robot that is shaped like a human? The human form is remarkably versatile, but there's a reason that industrial robots don't look even remotely human. Each has a specialised function and corresponding form that makes them far more efficient than a human being.

The best reason for having a robot that emulates human form and behaviour would be for interacting with other humans. There are fewer more basic human needs than that for sex, as much as we'd all like to be prudes and deny it. Though the uncanny valley may be an issue, human nature makes it unavoidable that this will happen eventually.

This was the key theme of only the second episode of the new cop drama Almost Human. I agree with you. However, I think there is a more immediate market in industry, and one that doesn't require nearly as complex robots.

Almost Human has been on my list of things to watch, I suppose I should move it up a notch. The industrial market will undoubtedly be served first by this (and Google's probably carefully avoiding any hint of what I suggested -- heck, I already got downvoted for merely mentioning the possibility), but I'm pretty sure it will happen. The ethical implications are interesting to say the least. Complexity is interesting, too. Historically, it's always been better to have specialised programming for the domain of use. Industrial/manufacturing purposes would indeed be a nicely delineated area. Once you make androids that have to deal with humans on a human level, pretty much all limits are gone as the entire gamut of human behaviour needs to be dealt with and a portion of it simulated. I suppose a sexbot of some kind is possible without being too complicated, just don't expect it to cook you dinner afterwards. Perhaps too much complexity would actually be detrimental in this case, as no sane person has yet worried about whether their blowup-doll likes them or has a good time. Moral and ethical issues will arise. But that hasn't stopped humanity in the past.

Highly specialised robots might be more efficient overall, but you lose the agility of being able to quickly give them tasks and repurpose them.

A lot of smaller businesses would take a general human shaped robot that they could visually and verbally give simple commands to (move all these boxes from here to over there, intuit that the stack needs to be stable), over some giant arm that they need to program/can't go outside/etc, even if it is more efficient at some specific set of tasks.

There's a reason Amazon are still using actual humans and barely organised heaps of unrelated products in a giant warehouse for a lot of their dispatch centres.

I'll grant you that, but why would such a robot need to actually look human? Legs can be useful in a few corner cases, but 3 or 4 of them would be more stable, or maybe just a compact set of wheels. Why does a robot need a face, or an area that looks like one? To do human work better than a human being (disregarding the never tired/complaining issue), a robot would need to be physically superior too. Strength can be such an advantage, but why stop there? Even as a human, I've wished that I had more than 2 hands, eyes in the back of my head, or simply a much greater range of motion in the ability to turn my waist and neck and other joints.

Seems to me that making robots look human is purely cosmetic and emotional, but limiting in functionality unless their function is to interact with humans.

espcially if your talking about moving things around a factory or store. I'm suprised forklifts aren't autmated already. they don't move fast I wouldn't think next gen kinect would be sufficent to detect obsticals find pallets. if you need something to go upstairs hand it off to something with flippers or legs. Also, I could see something like the assitant in Ironman with one arm and mobile being useful. Maybe some with suction cups as end.

I'm suprised forklifts aren't autmated already

They have been for a while.


> The best reason for having a robot that emulates human form and behaviour would be for interacting with other humans.

Alternatively, for interacting with things that were designed for humans, including most of the world we live in.

The Swedish series Real Humans explored this and many other related themes in great detail. I highly recommend it.


check out the robotic buttocks, http://youtu.be/vhHo6CUq4-o?t=2m45s. I love the researcher's creepy love hate relationship with the synthetic buttocks. WARNING: you can't unwatch it

For me this article tells me I am growing up. Previously I would have thought "gosh Google is getting into robots, well that sews up the market everyone will buy from them."

today I think - "Wow, Google is path blazing, there is going to be so many opportunities - let's dig out ROS and find a local company with a needs"

And you'll know you're grown up when you can spot a reactionary PR fluff piece instead of anything with actual content in it!

The Amazon announcement had a concrete end goal and time commitments, and was met with cynicism, whereas this fuzzy approach allowing people to project what they want into what they're reading creates a sort of delusional optimism among many.

I used to (and still do) run the opposite direction that the crowd is going. Its just that I am choosing my tribe - and my tribe is full of people who also run the opposite direction from the crowd.

So I have to be careful to spot both the larger crowd, the crowd that is now my tribe and when people are dropping FUD scent markers over their desired territory.

In short - robots still look like a hobby that will pay off the same way writing games for the ZX 80 or Atari will pay off.

> In short - robots still look like a hobby that will pay off the same way writing games for the ZX 80 or Atari will pay off.

You realize that that "pay off" is entirely social, right? For people who made careers out of game development, it wasn't because they made an early investment on the Atari. It was because they developed a passion.

If robots take off the way computer games did (and they probably will), the fact that you tinkered on them twenty years ago is not going to be good for anything other than stories over a beer, the same way my dad likes to talk about programming with punch cards. (Spoiler: tmk, my dad hasn't programmed anything in maybe 30 years. Kept up he has not.)

Comparing Robotics to Atari development is like comparing Computers to Atari development. It is the comparison of an idea and the application of an idea. You might be right about trendiness of robotics applications, but to throw the whole idea out with it is foolish.

Hmm, you think noisy drones flying around delivering packages within 30 minutes of ordering is a more concrete vision than a robot walking up to your door from a self-driving car? (last 30 ft problem).

I can easily imagine myself solving this problem in about 2 years if given access to Google street view data. Remember, Google probably has detailed (centimeter resolution) 3D maps of from their street view cars (they use LIDAR, afaik).

The Google plan is much more realistic than Amazon's announcement. IMHO. YMMV. I am not a robotics person etc.

Thanks for being a cynic! I am sure all big discoveries are made using concrete deadlines. I guess the quest for exploratory research is dead.

I feel like I'm greatly missing out on new stuff like this. I'm trying to get a turtle bot together and build a sort of lab component of Prof Thrun's robot car class. But I wonder what else I could be doing. Anyone have pointers?

It depends on your goals... if you want to get your feet wet with robotics development, a TurtleBot is a good place to start. You can learn the basics of ROS (Robot Operating System), play around with sensing (Kinect) for obstacle avoidance, mapping, and navigation. You can even build some pretty fun applications with a TurtleBot (eg. classic beer delivery).

For an extra couple $hundred, you can buy a decent hobby robot arm (eg. Robotis servos) and doing basic kinematics and manipulation. For the cost of a high-end computer, you can go pretty far these days!

EDIT: Also, I highly recommend reading Probabilistic Robotics. Not strictly necessary for many applications, but required reading if you want to understand most academic robotics topics.

It depends what you want to do. Do you simply want to get some experience in turtles or try to go for the rabbit? I think we are in front of a major disruption. So in order to get a good starting point in that domain I would look for very simple and cheap idea that can make you earn money. With that money you can then explore bolder ideas etc. So instead of searching for a cool idea, I would look for something usefull with value for people. I would look where there is still room for improvement.

Robotics is the next big thing because it will apply in so many fields. Not only robots, artificial intelligence as well. Better be the one who builds and conceive them than the one whose job has been taken oveer by them.

What are some good sites/blogs to follow to get more news like this?

IEEE Automaton Blog: http://spectrum.ieee.org/blog/automaton

RoboHub: http://robohub.org/

Hizook: http://www.hizook.com/

Everything Robotic: http://www.everything-robotic.com/

These are the main robot-centric ones (disclaimer: I run Hizook). There are lots of other websites that are mixed (robotics and other tech).

Some suggestions. To learn:

1) http://societyofrobots.com/ especially the forum

2) https://www.coursera.org/course/conrob

3) download the old coursera course on computer vision from some torrent site

4) anything and everything on 3d mathematics you can find.

Get a job making robots. Oh and the interview question is "how do you find a path for a non-point robot with non-line obstacles". The answer is to add the point-inversion of the robot to every point in the scene, and find a path in the resulting "configuration space" (google that and program it once. In 2d. No need to torture yourself with the 3d implementation). Worst-case followup question : and how do you find paths if rotation is allowed ? Tell the interviewer that you don't know, but he doesn't know that either (and it's easy to use the previous answer to come up with a -way to slow- algorithm suggestion, doing it efficiently and correctly however ...).

If you just want to make a robot. I suggest :

(shortcut if you got money to burn and don't care about electronics : http://www.robotis.com/xe/bioloid_en . Lego motors are slow and crappy)

1) forward kinematics (given 3 translated and rotated robot arm sections, what movement makes the end effector ?). Oh and mobile platform kinematics (robot arm is now a robot paw. What is the effect on a mobile platform if you move the arm, given that the end effector is on the ground)

2) make a robot arm. Bonus points if you make a walker.

3) actually program forward kinematics.

4) notice that the engines get really, really hot

5) learn control algorithms. Summary : the position of a motor is a function. Find the derivative of that function, speed. Make your control loop limit that speed (meaning speed needs to remain within -x and +x. Slow down if you find your robot risks exceeding that). Find the derivative of the speed function, acceleration. Make your control loop limit it's value.

5.5) find that the limits imposed in 5 are too strict. Find a way to relax them so that over a "small" time period of, say 2-3 seconds they're always true, but allow for small periods that exceed these limits.

6) notice that if you limit the torque (the derivative of the acceleration function to relatively low values) humans will not notice the movement of the robot, even though it doesn't preclude the robot moving quite fast. Have some fun with that.

6.5) attempt to make your robot pick up an unboiled egg. Break 100 eggs. Publish the traditional "look I can pick up an egg" (in 100 attempts you'll have a single egg that survives. Don't mention the other 99)

7) find ways to get your control loops based on other values. For example, find a way to have an end effector exert constant force on an object, as opposed to having a specific position. Find a way for your end effector to track an object.

7.5) make your robot pick up an unboiled egg. Get the breakage under 10%. Publish another paper. Do mention the 10 broken eggs. 8) find a way to do 7 while making sure your robot doesn't crash into itself, that generally there is no way to trick it into attempting to intersect with itself.

After step 7, you now know more than 80% of the people working on robotics. If you're still having fun, I'd definitely suggest getting into a phd program. Note that all this is bloody hard. Before you have step 8 covered you will be one of the 100 best-informed people in your state when it comes to 3d geometry and how forces affect objects. You will get irritated everytime you see a crane or bulldozer or escalator or elevator ... Your girlfriend will hate you for pointing out ways that those things can easily or suddenly accelerate large masses and cause disasters, and how easy it would be to prevent that.

Note that we really aren't that far when it comes to robotics. That means that with basic electronics you can achieve the state of the art (exception : battery life). It is not necessary to utilize expensive motors to achieve any of this, nor do you need things like a 3d printer or the like (servos + balsa wood(and something more solid, when you inevitable make a bigger robot) + saw + drill) will get you to step 6.5, and there is plenty of information on the internet on how to replace the servo ciruits to make 7+8 possible.

This is a fantastic list. I would like to add:

8) Robotics is a big industry. Find what interests you within it. It may be arms, rovers, drones, sensors, logic, etc. In my case, I'm interested in autonomous robots in the spirit of the google self driving projects.

I've been a robotics hobbyist for a few years and still can't answer any of those questions. I can get my turtlebot to follow me around and chase the cat though. Is there a practical programming robotics book/course out there? I've taken tons of math, controls, read robotics books, but still can't program anything harder than a PID or Kalman filter. I think part of that is me relying on ROS libraries too much.

Both of those can be really easy or really hard, depending. What exactly would you like to learn ?

Both things are related to control loops, so how about you make a slightly more complex robot that can detect it's position somehow, and get some pathfinding going ?

I also highly recommend bioloid (Dynamixel AX-12) servos. Next year there will also be slightly cheaper version (XL), but the current AX line is a great way for a software person to start working with robots right away.

You forgot inverse kinematics. If you want the robot to do something useful like pick up an object from somewhere and put it down somewhere else, forward kinematics alone won't help you.

A turtle bot? Kiel* would LOVE that!!

*Drow Tales reference...

In case the automated cars weren't enough of a wake up call, this is a pretty clear indication that the millions (billions?) of people we currently employ to do low-skill manual labor will be out of a job before too long. If we don't come up with useful for things for them to fill their time with, it's going to be a hugely destabilizing force in society.

What about a law that requires companies to train employees with a usable skill for X months before they can be replaced with a robot?

Hopefully all of these innovations will not only replace humans with robots but will also create new industries and new jobs.

Something that's more Utopian would be if there was no meaning of currency. I think it happened in Star Trek (never seen it but someone told me about the concept). If you have re-arrange the molecules of any object to create any other object, then why would everyone need to work? If you wanted MORE things then you could work, but to live day to day, would you have to pay for small items and food in the future or could you just "print" or create them from re-atomizing anything you already own?

To put some weight behind the quote from Andrew McAfee's quote (“The opportunity is massive,” said Andrew McAfee, a principal research scientist at the M.I.T. Center for Digital Business. “There are still people who walk around in factories and pick things up in distribution centers and work in the back rooms of grocery stores.”), just the external logistics of the Supply Chain of modern businesses accounts for around 7-9% of the GDP[1]. In developing countries, it is much higher (although I don't have a non-gated reference for other developing countries, the link in [1] shows China's cost at 18%). Interestingly, while inventory holding costs are included in that estimate, internal fulfillment costs (think Pick and Pack, Rail Shunt Yards, Crossdocking, etc) is not included as that is typically considered an operations cost. Those costs vary wildly by industry, but I can imagine it tacks on another 20-30% on top of that.

This would mean that the modern industrial supply chain costs the US around $1.5 trillion/yr.

[1] http://www.kpmg.com.cn/en/virtual_library/Property_Infrastru...

What kind of background is Google looking for when recruiting people to work on teams like these?

The article says they acquired 7 small companies recently, so I'm guessing they're looking for super-experienced people. Like people/teams who have designed and built a few robots from scratch before.

Bot & Dolly is hiring: http://www.botndolly.com/jobs

They're a good company to work for. Less nerdy than 99% of robotics companies.

I'm not sure how 'less nerdy' would be a plus for a robotics company, but good to have a positive review of the place nonetheless. Do you work there?

(tlb is Trevor Blackwell, a partner at YCombinator and founder/CEO of Anybots. It's safe to guess that he knows the field.)


Just guessing, but for me it's about the audience. Software in the 70s and 80s was very nerdy. With the rise of the web, it got less so, driven by the need to reach consumers and the influence of print design. In the last decade, a lot of tech is downright chic, and succeeds because of that. Look at the iPod and the iPhone, for example.

I think robotics is coming up on a similar transition. For years it was 99% research projects and industrial uses: pure nerdery. But Bot & Dolly is selling to Hollywood, and is very slickly marketed. Aesthetics are starting to really matter.

Whoa. It's worth checking out their video "Box":



We can see the two robots manipulating the rectangular screens, and infer one positioning the camera. What I'm trying to figure out, is if there are at least two additional robots positioning the projectors that light the screens, or if they just transformed the animation and the projectors are kept in basically the same place as the camera.

This is the most amazing thing I've seen in a really long time.

I don't know. In terms of excitement and creativity I think we're slowly catching up to the end of the 90s.

I don't work there, I just know them.

Bot & Dolly is half robot technology, half creative consultancy and they have a very cool office in SF.

Look at the speed, dexterity & flow of the robotic boom arm in this demo reel, WOW. Thats pretty much impossible for a human camera op to throw a camera about like that.


The question I've always had in regards to Bot & Dolly - how are they ensuring safety? Industrial robot arms are notoriously unforgiving and kill people every year. Putting untrained actors in close proximity with arms without some sort of safety system seems like madness. I hope I'm wrong though, since the company seems to be bringing robots to an interesting new place.

"Industrial robot arms are notoriously unforgiving and kill people every year"

um, how many people are actually killed by industrial robots? while the robots provide what economic value?

Only three reported deaths last time I checked. One in 1979, one in 1984 and another in 2009. Pretty sure more deaths occur from other equipment than that, every year even.

This is untrue, but it is hard to get international numbers.

In the US alone, it is about one death a year, purely by OSHA standards. This doesn't include things like "struck by robot, died three days later due to brain haemorrhage".

OSHA: https://www.osha.gov/pls/imis/AccidentSearch.search?acc_keyw...

Economist: 77 accidents with industrial robots in 2005 alone - http://www.economist.com/node/7001829

Research paper: 103 self reported and survey responded by 1995 - http://onlinelibrary.wiley.com/doi/10.1002/hfm.4530050304/ab...

Interesting, thanks. I went back and found the article I had read about this awhile back and I guess it was specific data about robot arms and probably wasn't entirely comprehensive:


Edit: one of those are a stretch calling it a "robot" A conveyor belt is a robot? Also, 33 results for robot, but 108 for Forklift and 11 for microwave, is it really all the much more dangerous than any other equipment? Should we ban microwaves (deliberate hyperbole, I realize those weren't deaths)?

I can't read the economist article "reached your reading limit" :(

77 accidents- I assume that's injuries, not deaths, right?

I would consider the OSHA count as incomplete for the reason you say- but I doubt there is more accurate data.

Right now I'd say the numbers look reasonable to not justify preventing the entertainment industry from using robot automation in movie making alongside actors (I think that's what was being suggested?).

Injuries. Death numbers are far harder to find.

The more pressing concern with putting actors in the mix is that they are necessarily untrained or poorly trained participants.

Most of these accidents happen with trained personnel.

I'm not saying that Bot and Dolly shouldn't do what they're doing. I'm just interested in what their safety procedure is to prevent accidents. "Don't get hit" is not a procedure.

Some. Some.

Both of these questions need more specificity - the former in the data available and the latter in what you're considering value.

Industrial Perception is hiring: http://www.industrial-perception.com/careers.html

they seem to have used a file photo from 2007 NYT story that has a toy robot. Hmm, bit tricky - to average person I think the photo suggests THIS is an example of the robots he's working on, no?


More likely he just was hard to schedule for a photo shoot, so they used a file photo and this was better than a headshot of him speaking at some conference. Unless there were new robots to show from Google, it wouldn't add much to take a new photo.

The long term plan of Google is "the skynet of terminator movie". Larry Page has always said it: http://blog.beacontechnologies.com/google-ai-artificial-inte... , http://www.artificialbrains.com/google .

Is it too early to reference Skynet and Terminator?

Why the down vote ? This is a pertinent question. It's a problem we will face soon or later. Internet has also it's problem. The ostrich attitude or obscurantism about them is plain stupid.

My answer to this question is that dynamite and machine have good use and bad use. There were also good reason to fear the bad use of them. But mankind manage to keep the balance largely in favor of the good use. I'll not prone obscurantism or the ostrich attitude regarding the risks of this new technologies. But we will have to manage that type of problem. It's easier when we have a little time left in front of us before things get out of control. I beleive that artificial intelligence reproducing the brain capabilities is just behind the corner. So we are in a race that I hope the good people and use will win.

I beleive that artificial intelligence reproducing the brain capabilities is just behind the corner.


All the cogs of the brain are well known and documented since years and their interoperation as well. But we still don't know how it all work together and what function is implemented by it.(I mean publicly. Some people may already know and keep it secret)

This is like DNA. All the puzzle pieces (molecules) were known but what was missing is how these pieces fit together. It was a aha moment that made the difference between before and after.

For the explanation on how the cortical neural network works the aha moment is the only missing step. All the puzzle pieces are known (of coures in the hypothesis that only bioelectrical mechanisms are in play).

> All the cogs of the brain are well known and documented since years and their interoperation as well. But we still don't know how it all work together and what function is implemented by it.(I mean publicly. Some people may already know and keep it secret)

I don't think there's a single correct thing about anything in this sentence, it's also a little bit conspiracy crazed.

What a troll! Anything more concrete to justify what you "think" or should I say believe ?

I don't see what could be "conspiracy crazed" in my comment. It looks like you believe that if someone knew how it worked, the whole world would immediatly know it as well. This might be a shock for you but this world is not sesame street.

No. It is exactly the right time. People really need to stop trying to bring scifi to life without considering that most of those scifi books/movies resulted in the extinction of mankind, or something near that.

People need to learn to stop generalizing from fictional evidence, especially wrt. to sci-fi tech.


Actually, when it comes to autonomous robots, I would generally say, "Holy shit we're all going to die unless these are very thoroughly regulated and controlled" is a pretty good heuristic.

A delivery drone that crashes into your car due to a software bug wasn't turning sapient and trying to kill you. You just ran off a bridge and died anyway.

> A delivery drone that crashes into your car due to a software bug wasn't turning sapient and trying to kill you. You just ran off a bridge and died anyway.

People crash into each other on the road all the time. I'd trust a machine much more than a human any day.

A thoroughly-tested autonomous driving program with all kinds of safety measures built in to handle things like firmware errors? Sure!

Some project that a three-man company hacked together in five months and are now selling without requisite safety measures and testing? Yikes!

The only job robot can't do is politician.

Not so sure ... read some Isaac Asimov. (And yes, I'm generalizing from fictional evidence, but I'm only playing).

Not keen on letting Amazon hog the PR spotlight, Google lets Andy Rubin out of the doghouse because the X Labs are all tapped out.

Could you explain how giving a guy free run of the place and acquiring 7 companies on his say so counts as "being in the doghouse".

Maybe i'm just not used to silicon valley doghouses; I know things are expensive out here.

Exactly. This would be a dream job for a heck of a lot of HN regulars: running your own shop, with no significant money worries, and a decent chance of changing the world. If this is the "doghouse," I'd be happy to be there too.

Just to state the obvious: Andy is not (and was not) in the doghouse. He had a massively successful run as the leader of Android and is now doing something else. I look forward to seeing what his team comes up with.

Or maybe this is what he went to X Labs to work on, and now they're moving forward with productizing it?

Why is he in the dog house?

Presumably: He bought Motorola. Meh patents, troubled OEM competitive with key Google partners. An expensive mistake even for Google.

No, it's all Google internal politics.

In my experience big SV companies have classes of staff on payroll merely to prevent them working somewhere else where they would represent a bigger threat to the core business. Google seems to have more than normal of this type of character.

Rubin is brilliant, and was absolutely the right guy to kick the carriers' asses and conquer the handset OEM world. I suppose that's all the more reason to keep him "inside the tent." I just think of it as horses for courses.

"Better to have him inside the tent pissing out, than outside pissing in."

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact