Hacker News Reader

Comment by lanevorockz | Hacker News Reader

I don't understand the scepticism, self-driving cars are already a reality and are working very efficiently in controlled environments. Anyone working on logistics already heavily relies on it. Another point, the "dedicated TPU" is also a myth. Neural Networks inference is very cheap and current nvidia hardware is able to run neural networks that would take years to train.

View on HN

None of the roads I drive on are controlled environments.

Highways are quite controlled, compared to city driving. You have separated traffic directions, no traffic allowed below certain speed. Markings are usually in good conditions and any works on the highways are marked and secured long distance before actual work place. I was just driving new car couple months ago with lane assist, front sensors and cruise control. Only thing I needed was that it overtakes trucks on its own and that is it.

That said people do stupid things on highways, but one step is to allow only autonomous cars to drive there and we are done. In the cities and rural areas that is not going to happen because costs would be too high.

I would be happy enough if I could get self driving car only for highways. If someone expects fully autonomous vehicles that can go everywhere, that is not something I think it will going to be available in next 100 years.

> Anyone working on logistics already heavily relies on it.

Care to elaborate? Sources?

AFAIK there are some self-driving cars, that follow guiding lines painted on the ground, already deployed in warehouses, production plants etc.

Industrial robotics applications are limited in scope and the environment is adjusted to make the problem easier than for autonomous driving on roads. Even then there are issues that mean often a human operator/supervisor is needed. General autonomous vehicles systems are not widely used in industry if at all.

Making a system that works on roads with other cars and pedestrians and varying quality of road markings and infrastructure is a different problem entirely. Tesla is trying to do this with a limited set of sensors and compute compared to others with makes it even less feasible. Waymo is by far the furthest ahead and they don't claim that full autonomy is close like Musk is claiming.

Reply on news.ycombinator.com