Godfather of Self-Driving Cars Says the Tech Is Going Nowhere | Hacker News Reader

Godfather of Self-Driving Cars Says the Tech Is Going Nowhere

futurism.com

28 pointsppjim25 comments

IMO, the only way this will work is if the network is autonomous, not the individual vehicle, and every vehicle will have to be connected to the network, no exceptions. I'm reminded of this as a Chandler, AZ resident where Waymo cars travel on nice straight, flat grids, with lane lines and easy line of sight with everything, versus what I observe spending time in New Jersey occasionally (no Waymo cars) where there are roundabouts, stop signs with obstructed views of the adjacent roads, hills, and everything else you can name that will overwhelm a vehicle's decision making capabilities. Then throw inclement weather for a good part of the year on top of it. If we as individuals want to share driving autonomy with the tech I don't believe it will work. We constantly break road rules that will ensure conflict on the road with self-driving vehicles.

What is an autonomous network? How does it recognize obstacles? How is latency and loss of signal handled?

1) That's called train.

2) Not gonna happen for normal roads, because it is worthless waste of resources and yet there is lot of additional variables, which you can't control, like people or animals entering the road.

That doesn't work unless you also eliminate all cars under direct human control, which is not practical, possible, or desirable.

And BTW, forcing adoption of automated cars only will quickly lead to situations like the one where you can lay down a chalk line around an automated car, and it winds up being immobilized to avoid crossing the lines. Far worse sabotage will ensue...

During my university years I used to bore everyone I could with this idea. I became quite obsessed with it and spent a lot of time working on simulations, the basic idea being that if we modelled a road network as a DAG, with intersections as vertices and "segments" (sections of a road between any two intersections) as edges, then the network itself could route traffic, allowing for balancing of load, priority of emergency vehicles, etc. An individual segment would be responsible for the transit of each of the vehicles on it, and each intersection would be responsible for transiting each vehicle through. The vehicles themselves would just be told what to do.

One of the main benefits I was interested in was the notion that an efficient network would likely mean that the speed of the vehicles could be brought right down to something pedestrian friendly.

Over the last few years I've watched with horror the version of autonomous vehicles that we seem to be getting, and I keep telling myself I'll dust off my old prototype and share it, but then paying the rent gets in the way (as an aside, the code has proven to be of some use - it forms the core of jsPlumb's Toolkit edition). I must admit to being increasingly skeptical of my original idea, though. As someone else mentioned, defining the interface between pedestrians and one of these autonomous vehicles is tricky. The easiest thing to do would be to completely ban pedestrians from the space used by the vehicles, which would be awful.

Like how Jay-walking is illegal right now

Yes, exactly. The story of how that came to pass is really infuriating.

He’s not The Godfather of self driving cars he’s a thief.

He tried to sell google’s self driving car division to Uber, but didn’t own the thing he was trying to sell. He walked off with all the tech and actively tried to take all the people. He was convicted and sentenced.

He’s also implicated in several other unethical self-enrichment schemes including founding companies to sell components to himself at a divisions he ran (multiple times), cheating his employees out of the ownership shares they deserved. He also tried to dodge court ordered reparations by founding a phony religion.

Calling him The Godfather of anything hides his crimes and makes the author complicit in his attempts to whitewash his reputation. Shame on you and him.

The title goes to Sebastien Thrun, who led the team that won the 2005 self driving car challenge, and went on to lead the Google self-driving car project. Levandowsky was Thrun's right hand man in the Google days.

I'm just going to repeat what I said on the last thread on this topic[0]. It's more or less exactly what I'd say to this article as well, modulo only the smallest tweaks.

---

As somebody who is a sort of self-driving car skeptic myself, even I have to say this kind of over-the-top skepticism is just as silly and absurd as the ridiculous self-driving car hype of the last 10 years or so. OK, so "we" have spent 10 years and $100B on this and it doesn't "fully" work yet. So what? Who said how long it was going to take or how much it was going to cost? This amounts to the little kid in the car screaming "are we there yet? Are we there yet?" The answer is "no", but we are going to get there eventually. That's pretty damn close to inevitable, barring nuclear armageddon or something.

So yeah, I'm a skeptic in terms of believing I won't be riding in a fully autonomous self-driving car in the next 10 years. Or maybe 20 years. But 50 years out? 100? 200? 1000? I mean, yeah, it's absolutely going to happen.

Remember the old saw "We always overestimate the change that will occur in the short term and underestimate the change that will occur in the long term. People overestimate what can be done in one year, and underestimate what can be done in ten."[1]

[0]: https://news.ycombinator.com/item?id=33168233

[1]: https://quoteinvestigator.com/2019/01/03/estimate/

"People overestimate what can be done in one year, and underestimate what can be done in ten." Yeah, it's definitely playing out with this technology. Remember just a few years back when they said there were babies being born that will never have to learn to drive. My daughter was born in 2016 and at this rate I'm fairly certain that in 10 years she will be going for her learner's permit.

> I won't be riding in a fully autonomous self-driving car in the next 10 years. Or maybe 20 years. But 50 years out? 100? 200? 1000? I mean, yeah, it's absolutely going to happen.

If it's not going to happen without major advances in technology, does it make sense for companies to be dumping billions of dollars into it today? Does it make since to continue endangering the public by beta testing these failed systems on our streets?

Sure, in 100 years it'll happen, but it is starting to look like we can save a lot of money, suffering, and energy by admitting where our current limitations are and backing off until technology advances a bit.

The only "benefit" I see in continuing now is that by the time cars that actually drive themselves are viable outside of limited/perfect conditions there will already have been plenty of lawsuits against car makers for accidents, deaths, hacks, and privacy issues. I'm not sure how well much of the laws and verdicts will fit with the systems we end up with in 100 years though or that it's ideal to constrain that future tech with the legal/regulatory fallout of today's failures.

You won't get major advances in technology without dumping billions into it.

How much of the current sensor technology being used was created by self-driving car companies? AI and computer vision systems are already being researched/developed in other industries, what else are these companies producing that hasn't been done before or isn't being done elsewhere?

Don’t quote me but I’m pretty sure the LIDAR sensor for most phone cameras exists entirely because of self driving cars. They drove the miniaturization of LIDAR sensors and the development of solid state LIDAR.

If it's not going to happen without major advances in technology, does it make sense for companies to be dumping billions of dollars into it today?

Good question. But it's up to those companies to decide. It's not my money, so I'm not going to try to tell them what to do.

Does it make since to continue endangering the public by beta testing these failed systems on our streets?

That's also a good question. At the very least, it needs to be made abundantly clear to anyone testing a self-driving car that somebody at the company will be held personally liable, up to and including possibly serving jail time, with regards to any damage / deaths caused by accidents.

Three things remain true about AI driving: 1) Artificial Intelligence has a LOOONG way to go to catch up with natural stupidity - even really stupid people are better drivers in the corner cases than most AIs. 2) The first 98-99% or so of AI driving is difficult, but fairly straightforward. The remaining 1-2% is both incredibly difficult, and worse, we have no idea how to do it. 3) Better technology and compute power will NOT solve this problem. I can point you to thousands of articles (both scholarly and popular) from the 1960s saying that we would have real AIs (intelligent machines) just as soon as we had faster computers and better tech. Well we have both: Not only that, but compute power has grown at rates that no one expected: a $75 Raspberry Pi 4 has more memory, more compute power, and more storage than a Cray supercomputer had in the early 1990s, and our new AI/ML GPU/NPU chips are many orders of magnitude faster than that. We have effectively infinite compute power - and yet the problem persists. We not only don't have Rosie the Robot - we don't even have robot floor cleaners that can avoid literally spreading unexpected sh*t all over the house...

> Anthony Levandowski

God father of self driving cars? That is a questionable statement

Seriously. The guy is a crook. I suppose if you mean godfather as in “The Godfather” the mob boss and criminal.

The way ahead is fairly clear.

First, self-driving sidewalk delivery robots: 4 mph. 50 lb: low-stakes.

Then bike lane delivery robots: 12 mph, 100 lb: medium stakes.

Then geofenced road vehicles: controlled stakes.

We're about at the comma after "First" above.

Meh, it’s not nowhere. Just drove through the big dig on the FSD beta. Was amazing to see it reduce driving stress by dealing with the constant traffic stops.

It’s called Full Self Driving, not Driving Assist.

Even if they never get anything else to work, it will have all have been worth it for self parallel parking.

I enjoy driving, so I don't want an AI to take the wheel, but the first time I had a car tell me "Yep, I can fit in there" and then park itself in the spot, was the most joyous tech-experience of my entire life.

I was in a good mood for the next few days.

Even if all progress stops, they have made the world a slightly better place.

Fundamentally they’re all solving the wrong problem. I don’t need self driving from my door to destination, I am happy if they just perfected (and really perfected!) FSD on the highway. Thats it, that’s the big win.

FWIW, I think this is already the case with standard Tesla (non FSD). I have a buddy with a 2 hour commute that does Duolingo while letting the car drive for the vast majority of his commute (which is mostly freeways).

Reply on news.ycombinator.com