Hacker News Reader

Comment by csours | Hacker News Reader

Disclaimer up front: I work for GM, this is solely my opinion and it is my opinion alone.

I think the tech press doesn't believe that Tesla is building a self-driving product.

I don't know if camera based computer vision by itself will ever be enough for full self driving. I think we can demonstrate that it is not enough at this time, and we can also demonstrate that human vision also fails at times; human vision has the benefit of some extremely specialized "software"

View on HN

"I don't know if camera based computer vision by itself will ever be enough for full self driving."

Rebuttal: If you had a 3D camera sitting in the driver's seat of a car and hooked up the control to a remote controller, I believe that car could be drive-able. A sufficiently advanced perception/control system could theoretically use the same 3D camera feed and replace the driver. Not saying it's easy, just that all the necessary info to drive a car safely can be contained in camera feeds.

A camera-based self-driving autopilot is also capable of driving without hesitation straight into an overturned truck on a sunny day less than a month ago:

https://www.autoblog.com/2020/06/01/video-tesla-model-3-cras...

I think it's a mistake to think that binocular or 3D vision is 'enough'. The visual cortex does a LOT of processing.

They don't seem to actually working toward L4 any time soon. It's just a steady drip of little features like "stops at stoplight" that are progress to fans, since none of the other self-driving companies are releasing anything, but what Waymo/Cruise/Zoox are stuck on (and have been working on for many many years) are much more nuanced problems that Tesla still have to face some day.

Having more data on the nuanced problems than any of the competitors is Tesla's most significant advantage, IMHO.

To me, it seems someone from Zoox wrote the article.

There are many tech and non-tech press which didn't believe in anything posted on the "master plans" published by Tesla, and they are still in denial.

Meanwhile, Tesla has been making huge leaps of progress and collecting billions of miles worth of data to improve their systems. What is available right now, and has been so for a couple of years, is unmatched by any of their competitors.

I am interested to see how Zoox and others will compete against massive amounts of data on corner case scenarios and real-world driving.

Tesla has millions of miles of experience with a level 2+ self driving system. Waymo has comparable experience with a level 4 system. Thus it makes no sense to claim that Tesla is closer to shipping a level 5 self-driving car than Google.

Billions of miles of actual data from real world scenarios. I am talking about the data they collect from the cameras and sensors, not how their system works or what level it is.

The data from real world is far more important than simulations.

Reply on news.ycombinator.com