Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Waymo One: The next step on our self-driving journey

medium.com

Post reply on HN
Print view View original

81–90 of 175 posts

Re: Waymo One: The next step on our self-driving journey

#81

Post by ghaff »

Hydraulix989 wrote: post #54
aantix wrote: post #45

Earlier quoted context omitted.

Curious; If you train a neural net to drive a car with X number of samples, and the car does something erratic, how do you determine which data item introduced the erratic behavior and how do you go about untraining that behavior?

Gradient descent is done in batches and over the aggregate collection of the training data so it's pretty lossy. Usually the error is more of the fault of the learned model itself, and then the culprit is attributed to neuron "mis"-activations post-hoc. Often, problems take the form of generalization error instead. And yes, all of these problems are still quite unsolved, which is why we don't have end-to-end neural n…

I think there's this popular perception that you just stream in lots of data from road vehicles and the driving just gets better over time--rather than continuing to discover individual edge cases more or less manually and updating rules based on those.

Re: Waymo One: The next step on our self-driving journey

#82

Post by lawrenceyan »

“Fares are based on time and distance, and customers can expect fares to be roughly on par with what you'd pay for an Uber or Lyft trip—perhaps even a bit lower. The above Waymo-provided screenshot shows a customer booking this trip, which is 4.6 miles long and takes about 12 minutes. Waymo charges $7.32 for the trip. I punched the same route into Lyft and Uber aps on Tuesday afternoon and got quotes of $8.29 and $9.38, respectively.”

Re: Waymo One: The next step on our self-driving journey

#83

Post by Tempest1981 »

One feature I like is the LCD displays for passengers, which (supposedly) show what the Waymo car is "thinking".

So if the car stops unexpectedly, the display might show, "waiting for a crossing pedestrian". Ideally with a simple map diagram of what's around the vehicle.

Re: Waymo One: The next step on our self-driving journey

#84

Post by a-wu »

I can’t wait for wheelchair accessible vehicles to be equipped with this technology. It looks like they’re using Chrysler Pacificas which already have wheelchair ramp conversions from VMI and Braun. Since the self driving equipment seems to be on the top of the vehicle, it seems trivial to mount it on a converted Pacifica. As a quadriplegic who doesn’t drive (yet), the possibilities with this excite me greatly.

Re: Waymo One: The next step on our self-driving journey

#85

Post by Judgmentality »

njarboe wrote: post #75
bryanlarsen wrote: post #53

Earlier quoted context omitted.

I'm a big fan of Sweden's Project Zero, one leg of which is setting and enforcing a speed limit of 30 km/h on any road or street without fully separated pedestrian & cycle infrastructure. I don't know if self-driving will help accomplish this everywhere, but I'd be ecstatic if it did. > We could even tolerate a little bit more errors because accidents would not be harmful too much. instead of "harmful too much" you c…

About the speed of a sprinting human, unsurprisingly.

The average human cannot sprint that fast. I'd peg the average person sprinting closer to half that (obviously there's a lot of variance, but I think only the top 1% of athletes could run 30 kph even for 5 seconds). For a frame of reference, try to run at the top speed of a treadmill (most max out at 15 mph, and most people can't go that fast - at least not at the gym).

Re: Waymo One: The next step on our self-driving journey

#86

Post by Quarrelsome »

bgentry wrote: post #78
rsync wrote: post #59

Earlier quoted context omitted.

"Keeping the safety drivers in the beginning, scaling up, and getting a year or more worth of actual real world use, before slowly scaling back the drivers is what I'd call a very responsible choice." A whole year! Truly sage-like wisdom from the MoveFastBreakThings generation of technical development. It is simultaneously comical, exasperating and sad to watch this tragedy unfold. We will slowly build our own, moder…

That’s funny, a month ago my car was totaled by a guy who fell asleep and blew through a red light. That happened at a right angle but only at 30mph. Since then I’ve been concerned at every intersection about what the other cars will do. I’m not sure how much of it can be blamed on cell phone use, but I swear that human drivers have been getting substantially worse over the last few years (at least in the Bay Area).…

While that OP didn't put it eloquently we are still creating a system where it would be technically possible to centralise the control of almost every single car at once.

We're going to do it but as with so many things we might want to ask ourselves if its wise to automate so very much while we still haven't aced security.

Re: Waymo One: The next step on our self-driving journey

#87

Post by munificent »

Tempest1981 wrote: post #83

One feature I like is the LCD displays for passengers, which (supposedly) show what the Waymo car is "thinking". So if the car stops unexpectedly, the display might show, "waiting for a crossing pedestrian". Ideally with a simple map diagram of what's around the vehicle.

This is both really cool, and really smart strategically.

I'm sure the Waymo folks know that one of the largest barriers to success for automated vehicles is public perception and fear. Each time a rider rides in one of these, they get a beautifully rendered window into the car's brain, which helps build the trust needed to sell the technology.

Re: Waymo One: The next step on our self-driving journey

#88

Post by munificent »

brianbreslin wrote: post #64

I hope they roll this out in Miami soon. Miami has some of the laxest AV regulations in the country. Also would happily pay to not drive in this mess every day.

I could be wrong, but a city with daily monsoon-like downpours part of the year and famously aggressive human drivers is probably not going to be an early adopter for AV tech.

Re: Waymo One: The next step on our self-driving journey

#89

Post by notahacker »

jacobmoe wrote: post #72
rsync wrote: post #59

Earlier quoted context omitted.

"Keeping the safety drivers in the beginning, scaling up, and getting a year or more worth of actual real world use, before slowly scaling back the drivers is what I'd call a very responsible choice." A whole year! Truly sage-like wisdom from the MoveFastBreakThings generation of technical development. It is simultaneously comical, exasperating and sad to watch this tragedy unfold. We will slowly build our own, moder…

You have it exactly backwards. You will _know_ what the vehicle will do because it will be programmed to do it. Meanwhile, that other vehicle could do anything because the driver might be texting, might have fallen asleep, might be in a fit of rage. Driving is an absurdly dangerous thing to do. We don't tolerate that level of risk in almost anything else in life. It just so happens that getting around at 60mph is so…

> You will _know_ what the vehicle will do because it will be programmed to do it.

I don't share either the OP's confidence in the attentiveness of human drivers or yours in the existence of bug-free complex software

The biggest differences between the two is that (i) software bugs can be potentially be systematic across all vehicle types and sudden, whereas human failure modes are discrete (ii) we have pretty good data on human failure modes which set the mileage interval for catastrophic failure surprisingly high, even higher if excluding human failures linked to behaviours society has decided not to tolerate, such as DUI and road racing.

Post reply on HN