Forums

Full Self Driving

Full Self Driving

Reliable Full Self Driving (FSD) is apparently way more difficult to achieve than Elon believes. News headlines today (9 Jul) are saying that Elon is not happy with his autonomous drive design team and has given them impossible timelines to meet goals. Consequently about eleven of the team members have left the company over the past several months. I wonder if we will ever experience reliable FSD with current maximum configuration of hardware, etc.

andy.connor.e | 2019年7月9日

I think our chances of getting reliable FSD with any other company is orders of magnitude less likely.

TabascoGuy | 2019年7月9日

1) So Elon's understanding of FSD and the difficulty level to implement it is apparent to you?
2) The restructuring of the Autopilot team was back in April.
3) Four members have left the team since then, not "about eleven". Four have been promoted.
4) Those that left were part of the path planning, perception, and simulation teams.
5) I wonder if people will ever give up on the lame attempts to spread FUD?

andy.connor.e | 2019年7月9日

@Tobasco

5) They'll never stop.

TabascoGuy | 2019年7月9日

That's the crazy part. It can't pay that well so they must actually believe their own nonsense. It just boggles the mind.

TeslaTap.com | 2019年7月9日

Tesla's engineers are some of the most sought after people, for those companies late to the game. Competitors are willing pay almost any amount to get a good engineer hoping it will speed there own FSD projects. No wonder a few engineers are willing to leave for the immediate payday.

Too bad hiring a few good engineers will never make up for incompetent overall design strategy and a tack of any meaningful real-world data. I expect many Tesla engineers stay at Tesla because they want to be in a team that succeeds and believe in Tesla's approach.

SamO | 2019年7月9日

Lidar is the future and it always will be.

andy.connor.e | 2019年7月9日

They dont have to believe it as long as one person believes it enough to not stop buying gas. Its just a game of making people believe in maintaining the status quo. One good example is when everyone said EVs were unreliable. No one wants their car to break down while driving it, thats an absolute nightmare, but not actually true.

TeslaTap.com | 2019年7月9日

@SamO - Best laugh of the day (and fully agree).

SamO | 2019年7月9日

@TT,

Stolen from Hydrogen, which I'm sure you recognized.

;-)

Here's a funny article from 2015 about how Hydrogen cars are finally here. I think they've sold under 10,000 in the last 5 years.

About as much use as retrofitting ICE cars to EV.

https://www.hydrogencarsnow.com/index.php/myths/hydrogen-is-the-fuel-of-...

Xerogas | 2019年7月9日

Can a LIDAR fan please explain how it can tell the difference between a red light and a green light at an intersection? And how it reads street signs? And how it improves AI regarding *understanding* its environment?

All the talk about cost is one thing, but what about basic functionality?

TeslaTap.com | 2019年7月9日

@SamO - 10,000? I think they would be ecstatic if they sold that many Hydrogen cars. I suspect it's below 1,000.

Ok checking, 2015 through end of 2018, Toyota sold about 3,000 Miria cars in California (the only state where it is sold). Hard to image that many suckers unless they were given away.

In Norway, after a H station blew up, all stations are shut down, so no fueling of any H car in Norway. Toyota and Hyundai both halted sales of fuel cell vehicles in the country.

@Xerogas - Yep, every LIDAR car has to have a vision system too - but every vision system doesn't need LIDAR. I just went to a sensors conference last week. The interesting sensor developments is radar. Radar already handles about 32 points, but units late in development have far more narrow "beams" that should eliminate any desire/need for LIDAR, at a fraction of the LIDAR price. Some of the demos I saw were more impressive than LIDARs I've seen, and radar can go behind opaque plastic (i.e. bumper) and is not sensitive to weather. I expect within 2 years LIDAR will be all but dead, except for those so invested they foolishly continue on (like Hydrogen cars).

blue adept | 2019年7月9日

@SamO

>>> "About as much use as retrofitting ICE cars to EV."

(•̀o•́)ง

tew ms us | 2019年7月10日

A couple of things:

The Mythical Man-Month
https://en.wikipedia.org/wiki/The_Mythical_Man-Month
Fred Brooks managed the development of IBM's OS/360; he found that adding more programmers didn't speed things up but slowed them down. The quality of the FSD team is more important than the head count.

Not an AI expert al all, but my impression is that scheduling the arrival of AI for FSD is sheer guesswork. You feed data and situations to the machine until it starts making correct decisions 99.99% of the time. It'll be a surprise.

Madatgascar | 2019年7月10日

I personally think FSD is a dangerous bet and needlessly endangers the much more important EV revolution. I know, Elon and the Tesla team are much smarter than I am, and I am sure they already have a system that works well enough that it seem to lie tantalizingly within reach. I hope it works out, but if I were placing bets, I would say it is more likely to be Elon’s Waterloo.

Consider:
1. There won’t be a million car robotaxi fleet, and there probably would not be critical mass for even a few cities’ worth for a long time. See (2) below.
2. The costs will probably be much higher than what we have been paying. Elon is now saying that cars with FSD are going to be worth “several times” more than non-FSD cars because of their earning potential. This may be true, but that doesn’t mean every FSD Tesla can suddenly be sold for that much more. I wanted FSD for myself and was willing to pay the current premium, but there is zero chance I would submit my car to a robotaxi fleet. The whole reason I drive a car instead of using public transport is to be able to go anywhere at any time, faster than an Uber. I suspect there are many others like me, and also many who will just have a problem with the abuse that comes from opening the doors to the general public. So if FSD enabled cars are validated but at a cost “several times” what we paid for our cars, only robotaxi participants will be able to afford them. This would appear to be a small minority, yet the hardware needs to go into every car?
3. The programming challenge is so obviously insurmountable. Just to take one example: How many taxi/ Uber rides have you taken where the driver is unable to drop you at the correct location, in a legal parking place, necessitating driver communications (“right here will be fine”), illegal stopping, and other split decisions? About a quarter maybe? How would FSD resolve?
4. The whole thing is at the mercy of regulators, regulators are controlled by politicians, and politicians are controlled to a large extent by an array of powerful interests who are determined to make Tesla fail. How likely is it things will play out in Tesla’s favor?

I hope I am wrong. Please tell me why I am wrong. I would really like to see it be a success, but wish the risk could be decoupled from Tesla’s overall execution risk.

TeslaTap.com | 2019年7月10日

@Madatgascar - Excellent points, and you could be right. Here's my view by point:

1) In 10 years, a million Robofleet seems very doable. Tesla will put out 350,000 Robotaxi capable cars this year alone. I expect they will be producing millions of cars a year in 10 years, so even if only 5% of owners elect to use their cars as a RoboTaxi, getting to a million seems easy. Now in 2 years, yes not so likely.

2) Costs are hard to say. Hardware costs are only likely to go down. Software costs, harder to estimate. I see the biggest risk is insurance and legal, which is really hard to estimate.

3) While the software develop FSD is difficult, it's far from insurmountable, unless you assume FSD will work in every possible situation. Yes, there will be situations FSD is not suitable - snow whiteout, lava flows, mud slides, etc. Just like humans, there are conditions to be avoided. There may be locations that must be avoided too, which can be geofenced. So if it works in 99.9% of locations and times people drive today, that seems very usable. For that last 0.1%, some cases can be instructed to use an alternate route to avoid the problem.

4) I expect FSD will be closer to Uber where regulations will be far behind what is actually done. There will be states or pocket areas where FSD will be locked out, but the pressure to allow it will be immense. Assuming one goal is to reduce traffic deaths, the federal government is more likely to favor FSD technology as it proves out. I do agree regulators may be the largest impediment. It will be interesting to see how it all works out.