Hacker News Reader

Comment by seibelj | Hacker News Reader

Is it really possible to have an "observer" of a self driving vehicle step in to save it in the event of an emergency? If you trust the vehicle for thousands of hours, and it never has a problem, you would of course start sleeping, reading, watching movies, etc. while it drives. The act of driving keeps you engaged, watching a car drive itself would be beyond boring.

View on HN

I'm not concerned about bugs that happen here and there. The occurrence is probably lower than human-made driving errors. The biggest fear is that since the control may be centralized, one failure or bug could potentially be pushed out to millions of cars, instead of being isolated to just one car in the case of human error

this is a great point. currently millions of individual errors are made by people that cause accidents. the future is 1 software exploited bug can cause millions of accidents. I am assuming overall software will have less fatalities but it would be interesting to see predictions on how the two would compare.

In the Otto model, no. The driver would be asleep or otherwise not at the controls. Operators for long-haul trucking are essential. Performing maintenance, safety checks, and loading/unloading assistance will have to be done by humans in the near future. I would not be surprised if this was something that was not automated away in my lifetime.

But these things can be cheaply localized into hubs and turned into shift work, can't they?

Yes. Some tasks (maintenance) can even be "shipped" to places where shift-work is inexpensive.

Or a single driver can command a convoy of trucks.

The goal seems to be fully autonomous driving for the easy parts (highway driving, which automatic systems are already quite good at) and a human driver just along for the ride to handle the hard part (city streets) at the ends.

Could there be the next step where the self driving ubers cars which just relay to and fro on the highway and human driven uber on the riskier areas inside the city?

Too cumbersome to switch cars probably for an end user..

For trucks the easy parts are also the dangerous parts, because those are where the accidents happen, because drivers fall asleep.

A driver could meet the truck at the freeway off-ramp to take over or supervise for the last few miles. Truck depots are already close to freeways, so depot-to-depot transfers could be completely unmanned.

A trucker's job would be to ride out to the edge of town and back again. Life would be like that of a UPS driver (though driving a big truck is real skill so it might command better pay). No more weeks away from home.

Today's truck stops could actually become where drivers start their drive. Do a local delivery, pick up new freight and drop the rig back off at the truck stop to send it on its way to wherever.

> Is it really possible to have an "observer" of a self driving vehicle step in to save it in the event of an emergency? If you trust the vehicle for thousands of hours, and it never has a problem, you would of course start sleeping, reading, watching movies, etc. while it drives. The act of driving keeps you engaged, watching a car drive itself would be beyond boring.

"BEEP! BEEP! Wake up Jim! Our simulation predicts we'll be getting into an accident after we turn the upcoming bend."

> "BEEP! BEEP! Wake up Jim! Our simulation predicts we'll be getting into an accident after we turn the upcoming bend."

This seems highly unrealistic. If the vehicle can run a simulation and know that it's going to get into an accident it can slow down, stop, pull over or simply avoid the accident all together.

No if there is an emergency it's going to be an "oh shit" moment and the trucker is going to have to grab the wheel and try to correct for a situation they're already in.

of course it's unrealistic. i think op was being facetious.

but, there is something to be said about our calling it "unrealistic" and not "impossible". :)

They certainly mention safety, but don't seem to suggest that the safety is due to the driver overriding the auto-pilot in emergencies. I do agree with your point, but they could be making the argument that the danger prevented from when the system is operating nominally will be much greater than the danger caused by the driver handoff in low confidence scenarios.

And even if the driver is totally alert. It's unclear to me why a human would be able to handle the situation that the machine couldn't. The only way I could see this making sense is if there were just certain conditions the car couldn't self drive in such as if it was too rainy.

It is a data flywheel. There are scenarios that the autopilot can't handle yet. So, having a human behind the wheel enables the transition between where we are now (computers driving 95% of the time) to where we need to get (computers driving 100%).

http://blog.algorithmia.com/machine-learning-trends-future-a...

I fear this when I was watching Highway Thru Hell.

Reply on news.ycombinator.com