San Jose, CA – Chris Urmson leads Google’s self-driving car program and is an adjunct professor at Carnegie Mellon University. Urmson was a keynote speaker at the 2013 Robo Business Symposium. Urmson talked about how Google’s self-driving vehicles hold the promise of saving lives and reshaping our relationship with the automobile. Urmson said when he first spoke to the car companies, as a representative for Google, “they would literally laugh at me.” They’re not laughing now, Urmson. Now all the OEMs and many auto suppliers have a self-driving project.
Urmson wouldn’t confirm whether they were working with one particular OEM, or were in talks with Tesla, but he did confirm Google has talked to all the OEMs.
Here are some of the questions asked to Urmson after the speech, and his answers:
Question: What does Google get out of this? Is what Google gets out of this is people spending more time in an internet connected world that allows Google to do more?
Answer: I think it is a possibility, but it is not a motivation. We look at transportation as a value inherent 2 trillion industry with a lot of opportunities to make it better.
Question: How do you get rid of the $70,000 lidar?
Answer: There are a lot of options for that that we are exploring.
Question: Who do you see as your biggest competitors in the OEM business?
Answer: Nissan, Tesla, General Motors, Mercedes has always been a leader.
Question: Is Google trying to set protocols for autonomous driving?
Answer: So, right now we’re not spending a lot of time on that, but there are amazing other communities that are working on that, car-2-car, connected car type stuff, and that will dovetail nicely into this.
Question: How much data has to be preloaded before a drive and do you require connectivity?
Answer: We give the car a map ahead of time, a map that covers the streets. We use connectivity but not for real-time control; we use it for things like traffic and construction zones.
Question: What is the range you use?
Answer: The Velodynes are at 10 HZ, the radars are at 15 HZ, and the cameras are lower than both of those two.
Question: Which will be more important in the future, radar, lidar or sensor?
Answer: They are going to be essential. They are very complimentary, a camera has a high resolution in color, the laser has medium resolution in direct range measurement and radar has low resolution in obscure temp readings. You blend the three, and you get a robust reading. It’s about complimentary performance.
Question: You haven’t created any of these products.
Answer: You’re right, we don’t go out and smelt aluminum, but we have relationships with all the tier one suppliers.
Question: Is there a part of your group that looks at neuroscience human adaptability or is it purely algorithmic?
Answer: We use a mixture, but we’re really not trying to emulate human behavior or human thinking. If you’re on the road, you want the other cars to behave like other vehicles. A classic example is if you’re driving down the freeway and your vehicle is next to an 18-wheeler even though the risk is not substantially greater than a normal vehicle, a human will move over and if you’re in one of these (autonomous) vehicles if they don’t do this, it feels incredibly unnatural. Even though it’s completely unreasonable it’s an important aspect.
Question: Is that an engineering or do you observe it from professional drivers?
Answer: We do a bit of both. We study how drivers work.
Question: Do you study the convoy driving?
Answer: No, we are not working on that. Volvo did amazing work there – there are a lot of advantages, but some disadvantages.
Question: Have you thought about how to alert a driver if they’re asleep etc., and the driver needs to take over if there is danger?
Answer: We’ve put a lot of thought into that. Adaptive cruise control is one way, Mercedes is doing the next step with their traffic jam assist. Mid-Level would be with some notice and then it could be that the car could just pull over. Because of texting, etc., driver monitoring is an active variable, and it is coming into the field. We are not actively working on it because others are working on it.
Question: Haven’t you had two accidents?
Answer: We got rear-ended once at a light, and one was Google’s fault when the driver was distracted behind the wheel.
Question: How many miles have you driven?
Answer: Over 1/2 a million.
Question: Is it a lot heavier?
Answer: Not substantially
Question: What do your kids think of autonomous driving?
Answer: They don’t see any difference. They get in the car, and someone else is already driving.
Question: Have you tested in more extreme conditions, where the roads aren’t visible?
Answer: So we haven’t pushed into that. We’ve done testing in gravel and drifting. Stanford did some great stuff on Pike Peaks. We haven’t tested in snow, and our current geometry wouldn’t work with snow. The dips off the side of the road, the trees are all white.
Question: What was on the Tartan (the Darpa vehicle Urmson was involved with) that is on a Google car?
Answer: The Velodyne laser.
Question: Have you looked at the new Velodyne laser?
Answer: I have looked at it.
Question: When an accident is unavoidable is there a system in place to minimize the damage to the vehicle?
Answer: So if it is unavoidable the best way is to take energy out of the system, applying the brakes, etc.
Question: If someone doesn’t have to drive is there any reason someone has to be in such a dangerous position as the front seat?
Answer: It turns out in the car the front seat is not bad, because the majority of the trips are single driver the majority of the safety regulations are centered around the front seat. but maybe if you have a self-driving car you have the ability to rethink the architecture of the vehicle.
Question: Will you have data that shows who was at fault in an accident?
Answer: Yes, but it’s less about who is at fault and more about what happened.
Question: Can you imagine how little insurance (on a car) you would need at that point (of autonomous vehicles)?
Answer: If you actually get to that point, it will be interesting to see how the insurance companies react because they are very actuarial, and they will understand the value to their customers. You would expect the price of insurance to go down.
Question: Is there talk of one lane?
Answer: I think it is tough initially because there aren’t any self-driving cars but over time if there is significant penetration there could be some discussion.
More comments from Chris Urmson:
Driving on a road is a matter of trust. If someone decides they are going to hit you, they are going to hit you. Practically these vehicles are going to be complex vehicles. If we get to this future state, it’s going to be really interesting, and some of the predictions about it are going to miss the mark. It’s more important that there is robustness in redundancy.
The problem with the automotive cycle is it is about 7 years, and a platform reboot is even longer than that and they’re not going to embed a new architecture in an unproven technology so what’s going to happen in the first generation of architecture you will have the architecture come in in something none critical like the entertainment system. That means you’ve got to look at the next generation of architecture. Frankly, they’ve got a lot at risk, and they’ve got a lot of risk-averse engineers.
Within four years (Governor Jerry Brown signed a bill last year that will regulate self-driving cars in California) Google’s Sergey (Brin) said they expect to have a semi or autonomous vehicle on the road.