Hacker News Reader

Comment by 1stop | Hacker News Reader

But it's still simple math. More weight, be more aggressive with brakes.

If human truck drivers can do it, then a computer truck with a 360 model of the environment around it can do it more efficiently right.

I don't understand the problem here really. Computers are way better at logical computation than humans and that's all we require here.

View on HN

Yes, but there are more problems to consider here. With some types of cargo, applying maximum brakes can actually damage it(think livestock for example). You also don't want the truck to be going into panic mode and applying maximum brakes for a plastic bag or a small animal. I'm afraid that our current technology can at best tell that there is something there, but it won't be able to tell what - humans at the moment are much better at visual recognition. It's just a concern at the moment, but I think it's an important one because it can lead to some serious accidents when the technology launches.

have you not been following the developments of self driving cars? Like the story of how google cars had problems gauging the intent of bike riders that were themselves stopped to let the car pass ... it looked like they were about to take off, so the car was too conservative and didn't move until the bike changed what they were doing. And then, once that condition was experienced and identified, it was fixed for all google cars in the entire fleet.

I'm pretty sure the car will be able to identify and plastic bag ... not to mention a simple cargo manifest that would apply parameters about how the cargo is to be dealt with.

My point is that all of these are "simple" problems. Oh the truck will identify a plastic bag. Oh the truck will have a cargo manifest and drive accordingly.

But these are all new technologies. Our visual recognition libraries are nowhere near ideal. I said this elsewhere but I will say it here too - our best of the best software can't tell a zebra and a sofa in a zebra print apart. I'm pretty sure that all that LIDAR is seeing is a spherical object on the road, it has no idea what it is.

The same with the cargo manifest - it sounds simple, but no one has a system like this. It would need to be made and implemented, and it takes time and effort - I'm not saying that it's impossible, far from it, but it's one more obstacle to overcome to make this possible.

> I said this elsewhere but I will say it here too - our best of the best software can't tell a zebra and a sofa in a zebra print apart.

Really? I find this surprising. Do you have any links or sources? I'd like to learn more.

There was a few articles about this exact issue on HN previously, a cursory search returns at least one of them:

http://rocknrollnerd.github.io/ml/2015/05/27/leopard-sofa.ht...

Fascinating, thanks for sharing!

Reply on news.ycombinator.com