[jhulla]

Out of curiosity, has anyone seen Google's self-driving cars testing down El Camino, University or Castro in PA and MV?

Each street presents uniquely complicated obstacles. In the case of El Camino, the police recently shut down the lights at an intersection due to Stanford game traffic. I'm curious if the self-driving cars know how to recognize hand signals from a legit police officer.

Over on University and Castro, there is a lot of foot traffic and sometimes people briefly double parking to drop off/pick up people. Some human drivers in these cases will (illegally?) cross the double yellow lines into the oncoming lane to pass the double-parked vehicles. Again, I wonder how the self-driving cars handle these scenarios.

[senko]

I'm also wondering about how portable the machine learning for the self-driving cars are.

Driving around, say, Europe, differs in subtle but important ways (no right on red, different on-pavement and side signaling, etc.) that it looks to me like they'd have to re-learn many of the aspects.

For example, Musk recently commented about how the Tesla car recognized what a valid parking spot was. That's totally dependent on location (country).

[qq66]

A self-driving car does not need to be able to handle every situation to still be useful. For example, if I had a self driving car that occasionally stopped and said "Please get in the drivers' seat, human driver required," it would still be a huge improvement over a normal car. And in the taxi fleet scenario, the cars could use alternate routes, or refuse a pickup in a place it can't get to and drop off as close as it can get.

[kasparsklavins]

By the time you are in the drivers seat, it will already be too late. A vehicle can't just stop and do nothing. A bad decision may be better than no decision at all.

[tiglionabbit]

It can in the scenarios listed above. If it sees the traffic lights blinking because a human is doing traffic control there, the car could safely stop and require a human to take over. If you're waiting behind someone who has stopped in the middle of the road to load or unload passengers, it's safe to just wait, or allow a human driver to figure out going around them.

[inimino]

That's not a self-driving car. A sometimes self-driving car is going to get people killed.

[qq66]

It has to be able to handle every situation safely, but doesn't need to be able to continue operating in every situation. If unexpected construction blocks the road, it needs to be able to stop and pull over, it can't go haywire and just accelerate right into a bulldozer. But it doesn't need to be able to navigate every situation. For example, a self-driving car which couldn't self-drive at night would still be hugely useful to me, I'd pay double for that feature.

[cheepin]

In my experience they don't really stand out other than being a bit slow (for which it actually got ticketed).

They seem to just be very conservative.

[mgraczyk]

The car did not actually get ticketed for driving too slow, it was only pulled over.

http://www.mercurynews.com/2015/11/12/mountain-view-google-s...

[aptwebapps]

I wonder how a fully-self driving car would handle being pulled over. Could be like Robocop but in reverse (i.e., human police escalating due to non-responsive robot).

[paganel]

> Some human drivers in these cases will (illegally?) cross the double yellow lines into the oncoming lane to pass the double-parked vehicles

This is legal in my country, I mean, crossing the continuous white line in case there's an obstacle blocking your advancement, in this case the obstacle being the parked cars you mention. Granted, there still are obtuse (I'd say stupid) cops in my country who'd not consider those parked cars as an obstacle and you'd get a ticket, and probably you'd get your license suspended, too.

[rmccue]

Where I am (Queensland, Australia), it's legal to cross the double white line to give cyclists clearance (1m for <60km, 1.5 for >60km). It's also legal to straddle the lane divider or drive on painted islands to do so.

I imagine the self-driving cars are going to have significant hurdles to adapt to different jurisdictions.

[visarga]

It would seem they need to test and adjust the car locally in each region before release. After they collect the data sets, they just need to finetune to the local dataset.