Hacker News new | ask | show | jobs
by GeorgeTirebiter 1593 days ago
An extremely wise set of decisions. I also see no reason, a priori, to 'blind' a vehicle to certain spectra of EM emissions; nor to accept that only passive sensing (cameras) can be used, when Active Sensing (probing, if you will) that Radar and LiDAR use is clearly giving the control computers more information about Reality (tm).

By using all (or almost all) available Active and Passive sensing technologies, fused with geofencing and operating at 'low-ish' speeds -- surely must be the fastest way to achieve 100% accident-proof self-driving vehicles that operate on ordinary city streets. Congratulations, Cruise. Keep up the Good Work.
2 comments
XorNot 1592 days ago
One argument would be that once you have many vehicles operating with LIDAR, it's unclear which systems are sufficiently robust against being disrupted by interference from other systems. Same with RADAR - while this is not new technology, we've never really had a regime with potentially dozens of systems operating in close proximity.

For all Tesla's problems, the automation-via-cameras solution is the one I find myself having the least problems with: using a single, obvious input (to humans), you don't wind up in a situation where you can have multiple differently-capable systems disagreeing on what they're seeing.

link
GeorgeTirebiter 1589 days ago
Generally, you solve this problem by using different (randomized) wavelengths, modulation (e.g. pulsed in a pattern), or if interference is inevitable, do something like WiFi or BLE does. It's not a big problem in practice.

I can only suggest you think harder about the 'all cameras' approach. Imagine say a snowstorm. Hail. Rain. Ice sheets. Sun in your 'eyes'. Cameras, basically, suck. Elon's game is to use suck-tech and make it un-suck with computons. Bad Choice.

link
wutbrodo 1592 days ago
> I also see no reason, a priori, to 'blind' a vehicle to certain spectra of EM emissions

The reason is simple: cost. The goal isn't to build a proof-of-concept safe AV, it's to build one that meets the safety bar _and_ is as cost-effective as possible, in a reasonable timeframe.

I happen to agree with the target-then-scale approach, but I also agree with Kyle that it's not a given that this is approach is definitely superior to the one that launches everywhere and tries to improve functionality.

link
bluGill 1592 days ago
Sure, but we already know that humans are not very good drivers (car accidents are #1 or #2 causes of death for age groups between 5 and 50). If you can do better than humans with more input, then that is compelling reason to use more input even if you can do just as good with a cheaper system.
link
wutbrodo 1592 days ago
I agree, but was narrowly addressing the claim " I also see no reason, a priori, to 'blind' a vehicle to certain spectra of EM emissions". Cost is the a priori reason, albeit one that is potentially balanced by others.
link