Startup Idea: Robot Cars

http://blog.kirigin.com/robot-cars

by aston • 12 years ago

99 92 12 years ago

blog.kirigin.com

pg• 12 years ago

There's a startup in the current batch using exactly this approach. In fact I'm pretty sure it's the friend he mentions at the end.

sown5• 12 years ago

Are they looking to hire specialists in the robotics field or would they take generalist?

That is, if I get my roomba create to use slam, particle filters, etc, as an engaged spectator, would that be OK?

Even if it's not, I'm doing it anyways. :p

Qworg5• 12 years ago

Any conversation at all about how to avoid the legal issues/being sued out of existence?

As much as I'd love robot cars to "move fast and break things", making an argument that "well, at least it doesn't kill as many people as teenage drivers" won't fly in court.

ivankirigin9• 12 years ago

In 2006, when I was chatting with the head of engineering at iRobot about this, he said: "this is what is going to make it work" and pulls out his car keys. Turning a system on will mean that the driver is making the choice. The same things were said about the first cruise control systems actually.

This is a dramatic over simplification. What will really matter is whether they work or not. From the performance I've seen, they will work.

It is also worth noting that the technology isn't in question, just the regulations. That means it is just a matter of when, not if.

Qworg5• 12 years ago

I'm currently working in a field that tries to lean on the same thing - the operator is making a choice.

It doesn't work, we still get sued, and money has to change hands.

I'd love to be laissez-faire about it, but it is a pretty critical issue.

iandanforth• 12 years ago

It is also worth noting that the technology isn't in question, just the regulations. That means it is just a matter of when, not if.

Hemp. Superior is many ways, illegal for a century and counting.

dm22• 12 years ago

Pressing a remote button telling the car to start seems no different, correct?

There is no reason that a fully autonomous fleet of cars is unsafe, only a matter of convincing stubborn old people to embrace advanced technologies.

If the car is unsure of what to do (road conditions change or map doesn't match road) then pull off to the side of the road and wait for a remote human to say "ok, you're clear, carry on".

thrill• 12 years ago

I believe that what will make it work is that boring old concept called insurance. Insurance on unproven concepts will be quite expensive - at least as long as our unlimited tort attitude exists in the US. However, if encouragement can be given to lawmakers to recognize that if any concept should allow limitations on damages then it should be where a machine is autonomously performing the activity.

josephpmay• 12 years ago

It probably won't be to hard for the startup to find an insurance company that's willing to partner with them. As long as there's a large payout in accident cases, I doubt civil suites would be a major problem. I personally would be more worried about criminal liability, which is best mediated by data collection. I could see early versions being "computer assisted" driving, where the driver is responsible (and liable) to take over if there is a problem. This is what Google is doing.

lazyjones• 12 years ago

I bet insurance companies will love the idea of self-driving cars - because there will be practically no avoidable accidents but drivers will likely accept at least normal insurance fees because they can't argue away the uncertainty associated with such new technology.

andrewfong• 12 years ago

I wonder what the capital cost for starting your own insurance company would be actually ... would VC money be sufficient for this sort of thing?

sown5• 12 years ago

Any conversation at all about how to avoid the legal issues/being sued out of existence?

I share your concern, but maybe just sort, uh, :) worry about that part later? It could prove to be fatal but it'll be neat to at least see?

I mean, at least quick enough to get a prototype together and then later-on the avionics level of code auditing would take place.

I've worked with former and current avionics engineers before, whom worked with Ada on fighter aircraft or civilian jets. It seems like a few lines of code per day is typical.

HeyLaughingBoy• 12 years ago

A point of reference here before the onslaught of "they must be crappy coders if they're that slow" remarks begins.

A few LOC/day rate is generally accepted to be the overall rate of development from Requirements through Final Verification/Validation. That is, you may write a product of 10kloc in just a few months of coding, but the entire software aspects of the project actually took about 2.5 years to complete.

I don't do avionics (though it sounds like more fun. Greener grass on that side, etc...), but this is the way we measure LOC in medical devices.

To your other point, well, that's easy to say until the people killed are your loved ones. No company that plans to stay in business (or its officers out of jail) can afford to be blase about killing its customers.

ivankirigin9• 12 years ago

I mentioned this at the end of the post, but if you want to work on a company building this right now, then just email me: ivan.kirigin@gmail.com

TaylorAlexander4• 12 years ago

Sent along my resume. Amusingly, I have just started actively looking for a job in robotic cars, and Google was the only place I knew of to apply!

therobot246• 12 years ago

there is also Telsa, Ford, GM, Toyota, Chrysler, Honda, ..., Oshkosh (not the clothes), John Deere, ..., and even the military - all of which are actively investing in robotic cars

sown5• 12 years ago

Thanks for offering to take contact info; I'm sure your inbox is exploding now!

ivankirigin9• 12 years ago

yep

Schwolop• 12 years ago

I do want to build autonomous vehicles, and I have the Ph.D. where I did just that to validate my ability - but, and this is a big but, I don't want to do it in Silicon Valley. There are plenty of excellent American robotics engineers available, but there are even more European, Australian, Japanese, Chinese, etc, who love their lifestyles and would rather work on something slightly less interesting to not have to move. The first hardware/software company that works out how to effectively handle remote teams and prototype hardware is going to dramatically increase their available pool of skilled workers. I for one would jump at the chance to work on something that compelling if I could do so from Sydney or Melbourne instead of California!

askar_yu• 12 years ago

>> "Public perception and policy-maker overreaction are real concerns, but as a startup, you don’t need to be as concerned"

I don't have any credentials to speak on the subject, but it seems that robotic cars is an area where traditional startup advices along the lines of "move fast" has to be taken very cautiously. I also believe that Google being paranoid about first robotic car killing someone is very much justified. Rightly so.

ivankirigin9• 12 years ago

I don't think Google can take any other approach, so we're agreed. My point is because Google is doing what it is doing, your company doesn't necessarily need to be pushing the regulatory front. This is great, because your small team can focus entirely on building a sufficient quality product to launch.

ufmace2• 12 years ago

I'm guessing that a big part of that is the combination of the potential for huge damages, and how big and public Google already is. One nasty accident followed by an ugly, drawn-out lawsuit could cause huge financial and PR damage to them. On the other hand, if you're a small start-up with no reputation to protect and not much capital to lose/go after, then worst case, you don't lose anything more than what's already been put into the single-purpose business.

HeyImAlex• 12 years ago

Worst case for you maybe, but it could set back the industry pretty significantly.

notahacker• 12 years ago

I'm not convinced lawmakers won't be convinced to go after individual company executives for reckless endangerment, negligence, manslaughter or whatever else they can dream up.

Especially if, Ford Pinto style, they've publicly stated that [for financial/reputational reasons] they shouldn't be as concerned about safety as another market participant, or that they should should aim to be "only just" better than fallible human drivers who frequently are held accountable for their fallibility.

I wouldn't want to be on the design team of a robotic car the first time it hits a child, even if the child acted in a reckless and unpredictable way that even a human mind would have been unlikely to anticipate and the software had demonstrated a far superior safety record to human drivers overall.

ForHackernews2• 12 years ago

Given how incompetent Silicon Valley startups appear to be at important-but-tricky things like data security[0], I'm pretty happy to not have to share the road with robot cars piloted by likes of Snapchat.

[0] http://www.nytimes.com/2014/03/03/technology/when-start-ups-...

TaylorAlexander4• 12 years ago

It seems like a leap to assume that engineers working on a product that could kill multiple people would be as careless as those working on mobile apps for sharing photos.

For self-driving cars, security and safety are features. For the others, for better or worse, they are afterthoughts.

ForHackernews2• 12 years ago

Plenty of apps like Snapchat and Secret position themselves as though privacy/security are their key selling points. But then they consistently fail to actually deliver on those features.

I would hope that engineers working on a self-driving car will do better. I guess we'll see.

zwieback• 12 years ago

I wonder if there are going to be local areas where robotic cars will work first, e.g. local areas like the Bay Area. It's hard to imagine lawmakers in 50 states getting to some kind of agreement. On the other hand, the interstate system might go first since it's already more unified.

The other thing I'm hoping for is that people realize that once they hand over the control to a robot they might as well give up individual transportation altogether and board a local robot minibus or whatever system ends up being most efficient.

ivankirigin9• 12 years ago

At the start, there will be drivers that are handing control to the system. So expect self driving cars on the road well before you get a bus without a driver.

The key thing about the interstates are their regularity. Small street roads are much harder. There are more spurious obstacles, including foot traffic, and perception is harder.

Even things like the ease of viewing lane lines matters. This is why solving a sub problem is so essential. If you wait to launch to be able to take on cobblestone streets of old london, a product that only works on interstate 101 and 280 in the bay area is going to launch first. They'll accumulate more data, and they'll win.

waterlesscloud2• 12 years ago

How do you imagine that working, a car that only works on interstates? It's manual on surface streets, then automatic on freeways?

sharemywin3• 12 years ago

why not automate semi-trucks?

ivankirigin9• 12 years ago

I mention elsewhere that at the start there will still be drivers in the cars. Semis will happen after commuters.

But yeah, long hauls were made for robots. iRobot's slogan was "dull dirty or dangerous". Dull counts for a lot to replace a human!

Qworg5• 12 years ago

Several major truck/auto manufacturers are working on the semi problem - it is a bigger focus in Europe than in the States right now.

Scania and Volvo both come to mind: http://www.bbc.com/future/story/20130409-robot-truck-platoon...

dm22• 12 years ago

There are billions to be made in the automated transport truck industry, there are large barriers to entry though.

Companies would embrace the technology because the cost of a driver is so high (100k+ per year to pay a driver, benefits, and insurance).

prawn2• 12 years ago

Or farms?

angusb2• 12 years ago

I liked this article, and this is only a small point in the context of how interesting the rest of it is, but it would be reckless to say that you need to exceed only 1M driving hours accident free to be better than humans. Of course this would lead to a lower empirical accidents:mile ratio for your new tech, but you still wouldn't have enough data to be confident that your accidents:mile ratio fairly represents the chances of the new tech causing crashes. I'm not well read enough on p-values/confidence intervals/chi-squared tests to explain why, so maybe someone who is can explain this if there's enough interest. Basically someone needs to get all Evan Miller on this (e.g. http://www.evanmiller.org/tesla-fires.html )

ivankirigin9• 12 years ago

The Tesla fire situation makes me the most worried here, because the facts on the ground don't matter. The press ran with the story because it was explosive. So even if you had a much more confident estimate, you think the press is going to grok your math?

The analysis I did was horribly incomplete, as I mentioned. The branding and marketing for the startup is going to matter a lot because of how emotional this whole robotic story can be.

angusb2• 12 years ago

Sadly you're right. But my point grapples with a larger issue too: should a startup launch if they aren't confident about the safety of their product?

Qworg5• 12 years ago

Not only that, but simply testing your system will not provide any sort of safety guarantee. Software errors are not just "show stopping bug" variety, but critical errors in odd corner cases that are not exercised until there's a fatal crash. See the Toyota unintended acceleration discussion: https://news.ycombinator.com/item?id=6636811

There's an entirely different way to write safety critical code that also specifies a process (IEC61508, MISRA C, etc). It isn't a guarantee, but it is required to have any sort of certainty in your system. This is before testing and validation.

samstave• 12 years ago

You know what I would love to see; a zone which is dedicated to ONLY pedestrians AND small robot cars (golf-cart-sized) which one can hop into, give a verbal destination and have it roboticize your ride to that location.

This would be great in, say, an area of Vegas to shuttle people around between casinos.

sharemywin3• 12 years ago

maybe you can pich it to disney.

dasickis3• 12 years ago

Check out Navia: http://induct-technology.com/en/products/navia, they'll be deploying in Stanford soon.

martythemaniak• 12 years ago

Ever since I started working on BarSense, I've been obsessed with computer vision (one of the most important subsets of robotics) and I think there'll be a wealth of products coming from that field.

One of the ideas I have is a type of smart dashcam - take the footage of one or more cameras (ideally 360°), apply CV and give the driver essential real-time info like cyclists or cars coming into the blind spot, sudden changes, lane drifting etc.

The hardware is cheap and easy to work with so the project is small enough for a startup, yet it's also something you can build upon and eventually scale into a full self-driving vehicle.

ah-• 12 years ago

This is already part of many modern cars, with the added feature that the car will also automatically break before you hit a cyclist. See for example http://www.euroncap.com/rewards/volkswagen_city_emergency_br...

avn2109• 12 years ago

The image of a car intentionally "breaking" rather than hitting a cyclist warms my bike-commuter's heart. Sadly I think you mean that the car will "brake." </pedantry>

nradov• 12 years ago

Subaru is already doing this. http://www.subaru.com/engineering/eyesight.html Some other manufacturers have similar technology. It has to be integrated directly into the vehicle to be of much use.

therobot246• 12 years ago

Don't write a check you can't cash.

A guy in my lab did something similar (not 360 degrees) - basically trying to do as much autonomy as possible via a single camera. His work is really great, however he must have spent at least 2 years on building a system to detect and remove rain drops and other distortions from the front window to get results that 'aren't terrible' for night and rainy conditions. All car manufacturers spend weeks out in the desert and up in the cold of alaska to get results against extreme conditions. CV will be no exception. There's a great paper (can't find it at the moment) that gives the problem of crash detection at a stop light. During the day..no problem, but at night, or rainy, or sunset (heavy glare), etc. CV just can't do the job, but what they used instead was a pairing microphone. The sound of a crash was distinct and easy enough to detect that it was a better solution to the problem in most conditions (not all, but more than CV could solve).

I'm a cv researcher, and I love to build and design algorithms with cameras. But saying your system will "eventually scale into a X" should never be uttered lightly, even if you have proof of concept on Y.

xinxian2• 12 years ago

Here is a nice list on what is already available: http://blog.a9t9.com/2014/02/self-driving-car-that-you-can-b... - yours is the collision avoidance system.

Having said that, I do think CV will create tons of novel products (if I only knew which ones)

TeeWEE• 12 years ago

Which startup robotic company are you talking about in the end of the post?

danmaz74• 12 years ago

It's in stealth mode I guess, or he would have mentioned it already.

therobot246• 12 years ago

This is much easier said than done. DARPA moved on because they work to push the edge of research into proof of concepts. A lot of good work has been done to make 'robotic cars' a reality (sqrt-slam, better velodynes, deformable parts model for pedestrian detection, etc), but definitely doesn't mean i would trust a small team of hackers with my safety on the road.

Even the Google team, as much progress as they've made, is like 11-12 people working full time for the last few years to build something worth demonstrating. However, despite their resources and expertise, still have a lot to overcome. In talking with Jiajun Zhu (manager of the Google car team), it became readily apparent that they still encounter situations where they are not fully confident in their system (such as on streets with too much brush/foliage). Given the rocky start (in terms of security, features, and stability) that most start-ups have, i would never trust their finished product.

MattHeard• 12 years ago

In talking with Jiajun Zhu (manager of the Google car team), it became readily apparent that they still encounter situations where they are fully confident in their system (such as on streets with too much brush/foliage).

Sorry, but did you mean to say that they still encounter situations where they are _not_ fully confident in their system?

therobot246• 12 years ago

yea, typo, thanks for pointing it out

dasickis3• 12 years ago

Also, Google's team (mostly from Carnegie or it's off-shoot teams) has been working on the problem for more than a decade as they took the Stanford car which came from Thrun who came from CM.

The problem to go after initially is best conditions highway driving. Additionally, low-speed (<=20MPH) is reasonable challenge. Where things get tricky is building out a model for medium speed (30 - 40MPH) where there's pedestrians.

http://www.youtube.com/watch?v=kOSvpPZzGpo

Someone• 12 years ago

"you need to make a product people can use as fast as possible."

Isn't this what many car companies already do? You have individual features, such as anti-lock braking (decades old), auto parking, lane departure warning and its improvement lane assist, brake assist, etc that get put into cars that aren't fully robotic yet.

Except for the PR angle, I still don't see what would make Google's approach a fundamentally better one. That may be because of lack of objective evaluations, though. Anybody know of any?

stackcollision• 12 years ago

I don't think robotic cars will ever happen, and I'll explain why.

Let's look at the recent MetroNorth railroad accident. Immediately afterwards, people were inevitably throwing around the idea of "why do we need train drivers at all?". The reason we need drivers is to satisfy a fundamental human misconception:

A while back, the DC Metro installed an automated train control system, with the intent of making the system independent of human operators. They flipped the trains onto automatic, and for a while everything ran fine. Then, one day, a sensor failed and one train plowed straight into the back of another at full speed while it was sitting in the station. That was the end of automatic train control in DC.[1]

Keep in mind that railroad traffic is a much more controlled environment than the open road.

Everyone knows that people make mistakes. We're (mostly) all engineers here, and we understand that technology is also flawed, like its creators. However, the public at large has this opinion that technology is magic; computers _don't_ make mistakes. The second that belief is proven wrong, people are immediately afraid of it.

Maybe one day a few robot cars will make it onto the road. But then someone is going to get killed by one. And despite these robotic cars having a stellar record up until their first accident, especially when compared to their human-operated counterparts, they will be taken off the road so fast your head will spin.

[1] http://en.wikipedia.org/wiki/June_2009_Washington_Metro_trai...

krapp• 12 years ago

And if it should happen that a human operator would have likely been able to avoid that fatality? Is it a mistake in all cases to doubt the validity of the premise that humans shouldn't be allowed to drive their own cars?

It is this loss of control which creates skepticism, not magical thinking about the nature of technology. Most people are surrounded by examples of technology not working, they know it's fallible, they would be afraid of autonomous cars for the right reasons - because the intent of autonomous cars is to deny them the ability not to trust technology blindly.

chuckwnelson• 12 years ago

I agree that people think computers are magic and as soon as one fails they get out their pitchforks, but I disagree that automated cars will never happen. It will happen just more slowly.

The changes will be gradual and they have already happened to a degree. Automatic brakes have already been implemented, more expansive GPS is happening. Next I could see an expansion on cruise control.

Its a combo between marketing new features and iterating on current technology.

I think the invention of the automated car will be a gradient, not a sudden occurrence.

RandallBrown2• 12 years ago

There are already cars today with adaptive cruise control and lane correction. On a highway, the car essentially drives itself. I imagine that with 10 or 20 years of incremental improvements we'll see something that's very close to a completely self driving car.

jrs99• 12 years ago

Wait. Are you saying 100,000 years from now, we will not have had robotic cars?

tegeek• 12 years ago

I think the word "never" should never be used for any problem that could be solved algorithmically.

There is no doubt about that we'll have robotic cars. But when, thats a million dollar question.

Houshalter3• 12 years ago

The thing is everyone says this. But I have yet to hear one person actually argue that "self-driving cars make mistakes and therefore should be banned". I think it's a strawman.

When the time comes to debate it, everyone is going to make the argument that they are still safer than human drivers. That wasn't true for the train operators.

prawn2• 12 years ago

You should probably phrase that as "Why people will resist the idea of robotic cars for some time" rather than "will never happen". Because they're fair points, but surmountable.

Are you offering 1000:1 odds, for example, on your "never" claim?

lnanek2• 12 years ago

He says Google isn't doing it right, then says to limit the product to things like clear lane markings and daylight. But auto manufacturers are already doing this. There are already products for auto-parallel parking and auto-following someone on a freeway and auto-applying breaks when someone is in front of you. I don't think it is a good idea to start your own company when you'd be better off trying to join examples of the two approaches that have much more resources behind them first.

TaylorAlexander4• 12 years ago

Those technologies are some of the pieces of a self driving car, but they are not self-driving cars. The idea is to deliver a car that can be programmed to drive a route from start to finish under its own control, even if it can only do so under certain circumstances.

When you can pull out your smartphone and call your car to you from the back of a parking lot, then get in and have it drive you home, then you have a self-driving car.

11thEarlOfMar• 12 years ago

A market that keeps coming back to me, don't laugh, is self driving RVs. A family of four could go to sleep in Chicago and wake up in Memphis. Spend the day tooling around Memphis, then hit the hay and wake up the next day in Miami.

No hotel costs, food costs could be the same as eating at home, no air fare or air travel hassle.

I suppose you'd have to wake up in the middle of the night to re-fuel, but hey, grab a midnight snack while you're at it.

It'd be the cheapest and safest way to tour a large area.

therobot246• 12 years ago

i like the way you think

This would be a dream - a railroad-esque type journey that is not at the wim of the location of pre-laid tracks

sown5• 12 years ago

This would be a dream - a railroad-esque type journey that is not at the wim of the location of pre-laid tracks

Not only that, but I could fulfill a life-long dream of solving a crime on a train. :D

Houshalter3• 12 years ago

That's interesting, but by far the largest market is trucks, taxis, and delivery. I don't know why everyone is focusing on consumers when it's basically just a convenience.

RandallBrown2• 12 years ago

Have you ever seen the Simpsons episode where Homer becomes a truck driver and it turns out that the trucks already drive themselves?

http://en.wikipedia.org/wiki/Maximum_Homerdrive

gregpilling2• 12 years ago

And the people who can afford a $500,000 diesel pusher RV can afford the self driving option. Who wants to park a 40' RV anyway? I am sure they would much rather check the $5000 (or even $10,000) dollar option.

dasickis3• 12 years ago

We have a forum to discuss open source implementations of robotic cars: http://www.sherecar.org/, we got pretty bogged down with some development we're working on.

chrisbennet• 12 years ago

I wonder if some sociopath will start a robotic startup that threatened to set back robotic car progress due to recklessness just so Google would buy it in order to shut it down.

therobot246• 12 years ago

I'm actually afraid that the 'stealth mode' startup that's being referred to in this thread will do just that. They brand themselves as the underdog team that took on Google's car with some baseline results. Then eager to bring in users they put prototypes on the road... unfortunately that's scary enough.

simonebrunozzi• 12 years ago

Evil, but smart idea. :)

ianbicking• 12 years ago

If I was going to enter with a strategy to go against the existing players, and I wanted to be more aggressive, I think I'd go with an fully unmanned autonomous vehicle with deliveries in mind, and design the vehicle to be as un-dangerous to bystanders as possible. You could limit delivery routes to only go in areas you feel confident about, you could do extreme breaking without any annoyance to passengers (who will be awfully annoyed if the car screeches to a halt because of overly conservative danger detection), you can change the form factor of the vehicle to do whatever is best for outward safety.

The downside is that you might not be able to piggyback on existing car regulations. And that could be a big downside.

truncs• 12 years ago

The author forgot about the fact that high quality/quantity data comes from good sensors. The cost of the LIDAR sensors (the one google using and this is the one which rotates on top of the car) is pretty high (around $50K AFAIK). This is one of the reasons Google is holding back in the hope that after few years the cost of such sensors would drop dramatically. Self driving car by just using CV just wouldn't work (atleast with the current state of the art).

gregpilling2• 12 years ago

I own a lift kit company www.traxda.com I want to build self driving trucks.

We are currently making 800-1000 lift kits per month. Lift kits are sold to people who spend a lot of money to have the coolest truck on the block. They would pay $5000 I am sure for self driving. We have manufacturing capacity and distribution.

I want to make the hardware, the actual retrofit kits. I have no skills at software, but my company is very good at making parts to fit truck suspensions.

Please put me in contact with anyone that you know interested in the same field.

Startup Idea: Robot Cars

Related Stories