Hacker News Reader

Comment by na85 | Hacker News Reader

> Given 1 or 2 more years of incremental improvements, your tesla would be capable of driving you most places I would think.

Except for anywhere in snow or rain.

View on HN

The improvements in autopilot in my friends model 3 is truly astonishing to me versus the first gen auto-pilot in the model s. If the incremental improvements over the last 2 years are any indication of what's possible over the next 2 years, we are going to see amazing things. It may not go everywhere in every condition, but so what? Isn't it still totally mind blowing how far it's already come?

>It may not go everywhere in every condition, but so what?

The "so what" is that it isn't really an automomous vehicle of it can't navigate the roads for a solid 50 % of the year in a good chunk of North America.

>Isn't it still totally mind blowing how far it's already come?

Not really, to be honest.

i don't know why you got downvoted. The engineer in charge at the latest autopilot meeting explicitely mentionned snow as not beeing specifically adressed yet. He did also mentioned that the autopilot performed quite well under snow right now, but that was indeed very surprising to hear that snow condition wasn't been specifically addressed yet, and yet they talk about robotaxi for next year.

They probably got downvoted for wrongly assuming the technology doesn't deal with rain or snow, and then presenting that assumption as fact. It works great in rain. Snow, I have less experience with, but it seemed OK during a winter vacation with snow.

With neural nets, a lot of things just get learned. You don't have to specifically address them to get great progress. I'm not saying snow is 100% solved, and yes specifically addressing aspects like snow in the future will help make it even better, but autopilot has worked pretty great so far.

was indeed very surprising to hear that snow condition wasn't been specifically addressed yet, and yet they talk about robotaxi for next year.

I've never yet seen a city with a taxi service which is reliable in heavy snow - things usually fall completely apart. (Not to mention buses, trains, planes, etc...)

Even if the robotaxis are shown to not be able to operate in snow, it's not necessarily an indictment of the whole robotaxi concept. If they can work in snow-free conditions, it's still covering the majority of time, in the majority of potential markets.

There's literally snow on the ground for the majority of the year in Edmonton. There's snow on the ground right now.

Hah, as I wrote my previous comment, inserting a caveat about Canada ran through my mind :)

My point still stands for most of the rest of the world, though.

Not on the roads though and that is where it matters. Most snowfall gets plowed up within a day of the snowfall ending.

Not in places like Winnipeg. Streets stay snowy for months.

Driving through snow and rain with the computer vision syste that Tesla uses, requires training its neural networks accordingly, which still has to be done. Driving through snow and rain with a system which requires LIDAR for operations, doesn't work.

This is something that Tesla fans don't seem to understand. LIDAR isn't an either-or proposition for anyone except Tesla. All existing LIDAR-based self-driving systems (Waymo, Uber, etc.) also use computer vision for reading signs and recognizing objects. LIDAR is used as a faster and more accurate way of determining distance and position in 3D space, with the visual data mapped into the 3d space, all in less time then it takes to finish the parallax-based positioning processing required by purely stereovisual systems like Tesla.

If you make the claim that Self driving cannot be done without LIDAR, then you are implying that if the LIDAR cannot provide accurate data due to weather, the car cannot drive autonomously.

Thanks, that is exactly the point I was trying to make. If you need LIDAR, you have a problem in bad weather.

Reply on news.ycombinator.com