Hacker News Reader

Comment by Animats | Hacker News Reader

Probably, so is Tesla's. Tesla's so-called "Full self driving" is still a level 2 system. It still needs constant driver attention for when it screws up.

"Unfortunately, we see that some automakers use the term 'self-driving' in an inaccurate way, giving consumers and the general public a false impression of the capabilities of driver assist (not fully autonomous) technology," - Waymo, which just stopped using the term "self-driving".

View on HN

Jumping from level 2 to 5 is going to be extremely difficult. I have no idea how we will ever accomplish that. Tesla is playing around a bit now, but this strategy will only work up until the moment a Tesla kills someone while provably being operated in "FSD" mode.

I don’t think you can hit a consumer price point with a vehicle with sufficient sensors. A taxi, though, can amortize the cost of a richer sensor suite.

Elon believes LiDAR is unnecessary and I’m inclined to believe him. After all, humans have exactly two “cameras”, while Teslas already have many more cameras, radar, and ultrasonic sensors.

To be fair, though, human eyes are so superior to any existing camera technology that it is hard to compare them. And the amount of processing attached to those eyes is staggering. LIDAR may be strictly unnecessary in the end, but for the foreseeable future computers can use all the help they can get.

> human eyes are so superior to any existing camera technology that it is hard to compare them

Not in every respect. Cameras can have a wider field of view, they can see wavelengths humans can’t, and of course you can have more than two, each looking in a different direction at the same time.

If Tesla can prove riding in their autonomous cars is an order of magnitude safer than human operated cars then it would be unethical to not allow them to switch on full autonomy.

It would be unethical that Tesla's magical actually full self-driving is only available as a autonomous system, not as driver assistance. You should not compare a modelS driven in favorable weather on favorable roads with younger people to the average car on the average road. You should also not compare a autonomous system with a system with no comparable driver assistance.

Tesla will convince me of driverless cars when a modelS driverless outperforms a modelS with a driver using emergency braking and lane keeping assist.

That's a big if. And as discussed recently on HN, it can't just be safer than average human drivers, it has to be safer than good human drivers. That is a pretty tall order.

Are “good” human drivers an order of magnitude better than “average” human drivers? Because that’s where I put my bar: marginally better than “good”, an order of magnitude better than “average”.

3 is attainable. Waymo, and probably Cruise, are at 3.

The problem with 3 and 4 isn’t that it’s hard but that it may not be a good idea at all. It’s a system you should monitor at all times but kind of works when you don’t? No security conscious manufacturer should make those.

Reply on news.ycombinator.com