California DMV Told Google Cars Still Need Steering Wheels

cartechboy writes Google showed us what it feels is the car of the future. It drives itself, it doesn't have a gas or brake pedal, and there's no steering wheel. But that last one might be an issue. Back in May California's Department of Motor Vehicles published safety guidelines aimed at manufacturers of self-driving vehicles. After seeing Google's self-driving car vision, the California DMV has told the company it needs to add all those things back to their traditional locations so that occupants can take "immediate physical control" of the vehicle if necessary. Don't for a second think this is a major setback for Google, as the prototypes unveiled weren't even close to production ready. While the DMV may loosen some of these restrictions in the future as well all become more comfortable with the idea of self-driving vehicles, there's no question when it comes down to the safety of those on the road.

Comments

Not surprising

[gurps_npc]

California is playing it safe. It will take a while for us to trust the software enough to remove the steering wheel.

In fact, it would not surprise at all if the brake itself is NEVER removed. I can easily foresee a situation where these vehicles are used to transport unwilling people, or simply undergo a malfunction and the occupant will always want the ability to stop the device.

But I can see the steering wheel and accelerator going away completely - don't want to let untrained people having the ability to make things worse.

Re:Not surprising

[TWX]

It's going to depend on who's allowed to use a self-driving car and under what conditions, and even so far as what seats are allowed to be occuppied.

I can see a tiered system where licensed drivers with a normal operator's permit are allowed to always occupy the driver's seat in a vehicle with the capability of full control. I could see a special provision of license for those who once held normal operators' permits that voluntarily gave up those licenses (elderly, poor vision, etc) so that they could basically pull-over the vehicle in a crisis. There could also be a special class of license for learners' permit operator licenses that allow the person to occupy that seat. Everyone else will be required to occupy any-other-seat unless all seats are occupied, then there would have to be conditions to allow that seat to be occupied while the controls are disabled.

Re:Not surprising

[jellomizer]

I agree, having a manual break should be required as a bare minimum.
Even if the software is perfect, if there is an unexpected power outage, you will need a manual break to stop the car that just may be aimlessly costing.

Re:Not surprising

[TWX]

As Toyota demonstrated to us, that manual break needs to be damn-near hardware-level too, or at least allow for an emergency override that interrupts the computer entirely if the main 'stop the car now' brake fails to work properly.

It would be terrifying to be in a self-driving runaway car without any controls whatsoever.

Re:Not surprising

[rossdee]

"I agree, having a manual break should be required as a bare minimum."

A manual brake would be even more useful, along with a kill switch for the engine.

Re:Not surprising

[pavon]

They may never be removed. Everyone is focused on the split-second decision scenario when talking about this issue, and on that I agree that humans will cause more problems than they solve. But there are many more situations where manual override is needed and beneficial. What happens when the car runs out of gas/charge and you need to push it to the side of the road out of traffic. Or the computer is malfunctioning somehow (software bug, squirrel chewed halfway through a wire, dead battery/alternator). Or when I need to move the car somewhere off-road that the AI refuses to recognize as a valid driving path. There are plenty of not so time critical scenarios where some sort of manual override is needed and those aren't going to go away even when we trust the software to do all the driving. Once we admit that they don't have to be intuitive for split-second reactions, then they don't have to retain the traditional layout, nor be designed for comfortable everyday use, but some sort of steering, brake control, and neutral gear select will always be needed.

Re:Not surprising

[Wootery]

Most "driverless car" situations involve a human with controls sitting on the other side of a radio signal connection...

Uh... no, no they most certainly do not. Where are you getting this?

Re:Not surprising

[TWX]

No, it wasn't, at least in all cases. There were definite computer control problems that led to the computer getting stuck in a mode where it had the throttle applied, and ignored the brake, shut-off, and gear-selector inputs because since logically the throttle was applied, those other inputs must be erroneous.

Re:Not surprising

[ShanghaiBill]

It's going to depend on who's allowed to use a self-driving car and under what conditions, and even so far as what seats are allowed to be occuppied.

Under California law, a licensed driver must be seated in the "driver's seat", and must be paying attention (no yacking on the cell phone). These requirements won't be permanent, but at least for the first few years of SDCs, that is how it will be. Once a safety record is established, and the pubic is more comfortable with the technology, the restrictions will be relaxed. In a decade or so, cars will likely be able to drive with no people on board, or even transport children with no adult in the car.

Re:Not surprising

[TWX]

Even in a muscle car, the brakes are more powerful than the engine.

That's complete and utter bullshit. I drive a '95 Impala. I can hold the brakes at a stop and floor the throttle, and cut the back tires loose. That car is completely bone-stock with factory-sized extra wide tires to give it as much traction as possible and I have enough power to overcome the hydraulic brakes at a dead stop. Hydraulic brakes don't have a chance if the car is actively under wide open throttle at speed, the brakes will heat up and effectively lose friction, causing them to fail.

By contrast, the e-brake on a car is not hydraulic, it's a physical steel cable that pulls on the rear brake's shoes to pull them into physical contact with the drums, or in the case of rear-disc cars, to have special auxiliary shoes make contact with the central hub of the rear rotor. Those brakes are designed to hold the vehicle still when parked on a slope so that if the parking pawl doesn't seat properly in the output shaft of the transmission, the car doesn't immediately roll down the hill.

As those e-brake cables are generally made of mild steel, they're prone to rust and fail, and it's very common to have to replace them in rustbelt climates. I've even seen them fail in desert, dry climates after enough years.

Stop spreading this crap that's completely false. Brakes on cars only really work right when there are no other forces acting at the same time. Throttle on? Brake will be ineffective. Going downhill at a high rate of speed? Brake will be less effective. Being pushed? Brake will be ineffective.

Re:Not surprising

[chis101]

I remember reading through that analysis a few months back. It was showed that a single flipped bit could cause the unintended acceleration. What was not shown was how this bit could possibly be flipped. So, it is far from proven that it was a software error.

Re:Not surprising

[ShanghaiBill]

How do you know to take sudden control of the car to save your life, or can you even possibly understand and react that fast?

It is best if you do NOT "take sudden control". In an emergency, the computer is going to have faster reaction times and is less likely to over compensate. In fact, I think that instead of requiring steering wheels and brakes, in a few years they will be banned from SDCs, as mounting evidence shows that most accidents occur when the human drivers "take control".

Re:Not surprising

[mspohr]

Some years ago I read an article about automatic control of airplanes which (somewhat facetiously) predicted that in the future airplanes would have one pilot and a dog. The dog's job would be to keep the pilot from touching the controls.
Humans are very erratic. Machines are predictable.

Of course

[Meneth]

Have they not seen "I, Robot" (2004)? Of course you need a manual override.

Backward-thinking by the DMV

[brunes69]

Any car that allows the driver to take "immediate physical control" makes the roads unsafer for all. The safest roads will be when ALL cars are autonomous. Having humans in the mix will just ruin all the gains that autonomous cars provide. Can a human wirelessly communicate with a car 5 miles ahead to know of a road condition and adjust it's speed in tandem with all the other cars in between to mitigate any and all danger in advance? Can a human react in sub-millisecond time to avoid obstacles thrown in their way. No and no.

Re:Backward-thinking by the DMV

[TWX]

Autonomous cars need to prove that they're capable of being safer than operator-driven cars. Right now they haven't done so, and until there's data there will be a need for autonomous cars to be manually operatable.

I expect to drive myself around for the next 30 years or more; I doubt self-driving cars in the price range that I can justify paying will come out any time soon.

Re:Horseless cars must accept horse harness

[HornWumpus]

Early cars were required to have a harness attachment point. Which was actually sane at the time. So is this.

Re:weakest link

[HornWumpus]

Eventually is a nice word. You can be completely wrong today but adding that one word...

Star Trek

[Barlo_Mung_42]

If there's one lesson I learned from Star Trek it's that you always, ALWAYS, include a manual override.

Never gonna work ...

[gstoddart]

California DMV has told the company it needs to add all those things back to their traditional locations so that occupants can take "immediate physical control" of the vehicle if necessary

The transition time from the computer giving up to the user having to take control is always going to mean this is impossible.

If you're reading the newspaper, you are not going to be able to transition to operating the vehicle in the event the computer gives up and says it's all up to you.

I've been saying for a while, that a driverless car needs to be 100% hands off for the people in the car, or serves no value at all other than as a gimmick.

I will believe driverless cars are ready for prime time when I can stumble out of a pub, crawl into the back seat and tell the car to take me home. Anything less than that is a giant failure of automation waiting to happen, and a convenient way of dodging liability by pretending that users are expected to be in control of the car even while the AI is driving.

As long as there is a pretense of handing back to the driver in even of an emergency, this is a glorified cruise control, and I'll bloody well drive myself.

If I'm ultimately responsible for the vehicle, I'll stay in control of the vehicle. Because if there's a 10 second lag between when the computer throws up its hands and says "I have no idea" and when the user is actually aware enough and in control, that is the window where Really Bad Things will happen.

Move a broken down vehicle?

[HockeyPuck]

If a driverless car has no manual means of steering, and if it broke down and you had to push it, how could you control it?

Short term

[ZombieBraintrust]

That is very short term thinking. 15 years after driverless cars are released your going to have a whole generation of people who never learned to drive a car. People who get the license and then forget everything because they haven't touched the wheel in 10 years. There is not going to be anyone quilified to drive a car. You will just have millions of amatuers with a skill level of a 16 year old.

How does it handle the unexpected?

[wcrowe]

I like the idea of a self-driving car, but I still don't understand how the self-driving car finds a parking space, or gets eased into place in the garage for maintenance. How does it find it's way around an unexpected hazard, like a downed limb, or washed-out area of the road? How does the self-driving car know that the road is flooded or otherwise undriveable? How does it know that the power is out at an intersection that normally has traffic lights?

As usual, the wrong question

[taustin]

That Google thinks their self driving cars are ready for the open road isn't the issue. The issue is that they think they are ready to go straight from traditional cars to cars with no ability for the human passenger to take control if the new, unproven technology fails. That, by itself, convinces me that Google's judgment is flawed, and cannot be trusted. Were I making this decision, I wouldn't let Google's cars on public roads at all until they show some evidence that they understand why this is a bad idea.