Hacker News Reader

Comment by Reedx | Hacker News Reader

Perhaps, but you can't dispute that humans don't have LIDAR.

It's just a matter of time before AI can drive better than humans even if using similar vision. Especially in aggregate (e.g., factoring in distracted drivers).

View on HN

Humans don't only have vision. Humans have an ocular-vestibular system that makes it possible to perform SLAM[1]-like processing; when in motion, your balance and visual input like the parallax effect can help the brain build a mapped 3D volume and determine your position, orientation and velocity and your relation to other objects within that volume.

VSLAM, visual SLAM, which is what you have to employ if relying largely on vision-only, is so far cruder than that.

I'm sure Tesla has some sensor fusion that also employs GPS and short-range radar/ultrasonar to help.

You can get some velocity and positional data through GPS, but it will have quite a lot of latency and is very coarse (at least the commercially accessible part).

SLAM is much, much easier to do with LIDAR since that will give you a point cloud 3D model. With VSLAM, you have to infer the depth data and build it up over time, and it's a much harder problem to solve reliably. OTOH, LIDAR's Achilles heel is, however, heavy rain/sleet/snow. Waymo has shown off using deep learning to filter the bulk of the noise out, but it looks like it's still early days there.

Vision isn't great for sleet/snow either, to be fair.

[1]: simultaneous localization and mapping

Tesla had a vid of their system doing that quite well https://youtu.be/Ucp0TTmvqOE?t=8324

Every smartphone has a gyroscope and accelerometer. Surely Teslas have them too, feeding into the model with the other sensors.

This post seemed to use a lot of acronyms and technical words to say not much, are you just saying that humans have more than one camera and gyros, integrated well?

Sure, but they have a vastly superior processing unit to make sense of those images. The question isn't necessarily if binocular vision is better than LIDAR or not but rather which one works best with the current crop of artificial brains and algorithms.

I am not considering here situations where either system is impaired since that will lead to varied degrees of failure with both.

True, but lots of animals with much smaller visual cortexes also manage to avoid running into things.

This is a good point, but it ignores the often forgotten other half of the equation - human (or in this case animal) eyes. The human eye has orders of magnitude more dynamic range than the best DSLR cameras. This means in edge cases the camera will not see something important, such as a person, when a human will. And when I say the camera will not see it, I mean even a person looking at the camera will not see it, because the image is obscured or washed out or something similar.

Garbage in garbage out. Don't skimp on your sensors.

You can get arbitrarily high dynamic range by tinting some pixels, or using a second camera at a different iso. As a silly example, imagine combining the image from a low light camera with one sitting behind a welding shield.

And sure enough their brains are more advanced and capable than our current computers, especially the ones that easily fit in a car. Even human vision (for the purpose of driving) is more advanced than cameras you might see in a car.

And you're oversimplifying a little, it goes far beyond just avoiding obstacles. You need to be able to anticipate, learn and adapt to completely new situations, understand consequences and things that aren't strictly related to what happens on the road. That's something even most mammals fail to do. Our computers are not there and they won't be for some time.

But not when moving at highway speeds with many other moving objects.

Small birds chase each other around, fly through thick forests or jungle, dodge predators including larger birds, and fly in large flocks. Based on the squabbles I see from my front porch, they seem to at least match human drivers on reaction time.

I'm not convinced absolute speed is that important of a variable, when comparing to something much smaller dealing with closer distances, but some birds actually do reach highway speeds and beyond: https://www.worldatlas.com/articles/the-fastest-birds-in-the...

But the fastest relative to size might be the hummingbird, which can fly 30mph backwards: https://www.ponderweasel.com/what-bird-can-fly-backwards/

The Anna's hummingbird has been clocked at up to 50mph in a dive, or 385 body lengths per second. https://www.wired.co.uk/article/the-hummingbird-thats-faster...

I have also seen birds fly into buildings, cars, planes and windows.

Good point but in fairness to birds, if there were a giant sheet of glass blocking the road then humans would probably crash into it too. They also might find it hard to avoid a 747 flying across the road.

Humans also have an unacceptable accident rate. The goal is to far surpass currently-accepted safety standards, and building a 3D model of the world at 10+ HZ can only help with that.

>Humans also have an unacceptable accident rate

How is it considered unacceptable? It has been acceptable for humans to drive in every country since 100+ years.

Anyway, the biggest problems with humans driving that cause accidents are distractions like smartphone usage, sleepiness, excessive speeding, tailgating, road rage, drunk driving, old age, being new to driving, being tired, medical problems like strokes, seizures, diabetic comas etc. which automated driving will solve.

Reply on news.ycombinator.com