Is Tesla Autopilot Way Ahead Of Rivals In Self-Driving Race? Video

MAR 14 2019 BY STEVEN LOVEDAY 48

What makes Tesla Autopilot the clear leader?

Sure, Tesla Autopilot is still a “hands-on” system. To say that there haven’t been problems caused by the technology would be a lie. However, we don’t see all the times the technology has helped a driver or even saved a life. There are a handful of other companies that are seriously working on self-driving technology. However, Tesla is at the forefront in terms of having the feature available in its entire fleet of public vehicles, testing it over an incredible number of miles, and improving it on a regular basis via over-the-air software updates.

It’s essentially impossible to know if Tesla’s semi-autonomous tech is “better” than that of Cruise Automation, Waymo, and others. As far as being ahead, at least in terms of application, testing miles, and real-world, feedback-driven improvements, Tesla is really the only company that is dominating the market.

Will we continue to learn about problems? Of course. However, if you take into account the exponential amount of beta miles driven in Tesla vehicles all over the world, the significant problems are really few and far between. In many cases, the worst problems have occurred when a driver isn’t paying attention or taking over when needed.

Ben Sullins (Teslanomics) dives deeper into the situation and data. Check out the video and let us know your thoughts in the comment section below.

Video Description via Teslanomics with Ben Sullins on YouTube:

Is Tesla Autopilot 100X Further in Self-Driving Race?

In a recent interview, Elon Musk revealed just how they collect feedback from drivers to improve their self-driving algorithms. Let’s see just how far ahead they really are in this analysis.

Categories: Tesla, Videos

Tags: , , , ,

Leave a Reply

48 Comments on "Is Tesla Autopilot Way Ahead Of Rivals In Self-Driving Race? Video"

newest oldest most voted

There was a scary tweet yesterday that was showing what the AP was seeing, and at one point the AP was detecting/identifying a truck turning and its trailer but still wanted to get under trailer (green path)

A green path indicates a driveable area like a road, not necessarily where it wants to go. Those videos are also from people who have ethically hacked into the system to see how it works, it’s not from Tesla and may not show a complete picture of what Autopilot is actually seeing or doing or deciding.

Navigant Research’s annual ratings of autonomous driving systems were just released. Once again,Tesla is rated as bottom of the pack.

https://amp.freep.com/amp/3142974002

That is because they are clueless (as always, their predictions on batteries weren’t any better).

AI is based on training, and training in cars is based on how many miles you put on the cars. While others try to start out at level 3-5 with a few cars, their training data is very limited.

Tesla’s approach is to start from level 2 and move up to level 3+ through training their AI. By having millions of cars on the road training the data, they have one of the biggest advantage on self driving.

The way that everyone else is developing their technologies, is to do billions of miles in simulations of every conceivable interation of real-wold situations and develop deep learning solutions for them.

Tesla are doing something fundamentally different to many of the other manufacturers in that list. Companies like Waymo are starting with the far more complex town and residential situation, which can then be easily adapted to the relatively easy highway situations*. Tesla (and most of the current Level 2 systems available to customers) are designed to work with the much easier highway situation but will take a lot of adapting to work in more crowded, complex locations like the residential streets.

In this way it may seem that Tesla are way ahead, but only in as much as they are are generally better than what is available to consumers right now, for easy low hanging fruit.

*Highways are simpler to drive, with much less cross traffic, pedestrians, wider lanes and simpler road systems.

I’m not denying that streets are harder than highway, (though highways do have some parts more difficult than streets such as high speed car cutting you off vs a low speed city driving), but Waymo can only put so many miles on their cars. Tesla at this point probably puts on as much miles on autopilot in 1 day as Waymo does in a lifetime.

This is why Waymo is looking to start a taxi service going, so that they can get more miles under their belt.

Waymo use simulated miles instead, which generally work very well.

And they aren’t starting a taxi service so they can get more miles under their belt. Their business model is autonomous taxis… They’re building an autonomous system so they can roll out autonomous taxis.

Ford is also working with this business model – the main aim for their autonomous tech is to build a large autonomous fleet. Presumably in the future it’ll trickle down to consumer vehicles when they believe it’s infallible enough for consumers to use safely.

Realistically they’re totally different strategies to Tesla, who are running a consumer model, rather than a fleet model. They will be far more likely to have a driver in the vehicle for the next decade that can take over for the difficult stuff. Waymo and Ford won’t.

And yes, I know Tesla have this idea of eventually letting people rent their cars out as taxis, but realistically that’s a long way in the future (especially if LIDAR becomes a requirement) and most Tesla owners won’t be doing this.

Tesla also uses simulated miles, but that isn’t enough. Simulated miles can account for general scenarios, but not niche scenarios. And getting through all those niches is the true challenge to self driving cars.

Simulations are created by humans

You have it wrong. Tesla is not designing their system for highways, they are simply collecting data from all Tesla’s all the time. Every move any Tesla makes is fed back to a central database and there is an AI system that analyses the data to learn. Where others teach their system to drive, Tesla’s AI learns on it’s own and will continue to improve the more miles that are driven. When a human corrects something the Autopilot is doing the AI is using that to learn.

That survey is about the potential future of fully autonomous driving, of which none are available in the market. It has almost nothing to do with currently available *driver assistance* features like Autopilot.

BTW industry funded “research” tends to (unsurprisingly) have built-in bias. It reminds me of the Gartner study that predicted Microsoft Windows Phone would overtake iOS worldwide back in 2017. I’ll give you one guess who funded that study.

So, self driving is a big, challenging, problem to solve. Software, deep learning, computer vision, sensor hardware, auto engineering. Tesla has taken the stance that they don’t need LIDAR, like everything they do its a big risk, but if it pays off they will leapfrog all these companies that have spent billions driving with $70,000 sensors on minivans. Whoever gets it out the door first, I’ll be buying it.

The biggest conference for advances in Software, deep learning, computer vision, sensor hardware, auto engineering is next week in San Jose, and guess who isn’t even there? Tesla, that’s who.

Any fully self-driving car will have to be equipped with either lidar or phased-array radar, or both. Tesla is denying reality pretty hard in claiming they can do everything with visible light cameras, which can’t see in the dark any better than you or I can. And please, don’t argue in favor of infrared cameras; they are expensive and require active cooling. (Solid state lidar is now considerably cheaper.) Also, once you have to use two different sensor systems and two different software systems to interpret those images, then what advantage could there possibly be to trying to rely on visible light cameras as the primary sensors?

You will be buying a Tesla then since the others are all for commercial taxis and not for sale to you.

The article stated that most of the leaders are using closed tracks and other limited variable situations. Conversely, Tesla received a low rating due to “high profile crashes” in the real world. So whether you think it’s a good or bad idea that every Tesla on the road is beta testing new Autopilot features, the fact that Tesla is getting their data from real world (instead of controlled experimental) conditions means they have the opportunity to grow from a limited system to a more full featured system faster. The multiple corner cases Mr. Musk referenced during the interview in the video are all opportunities for Tesla to improve that the others do not have. So Tesla has a clear data advantage. The question then becomes: how fast and effectively can they use that data to improve the system?

Right. Using simulated driving to advance self-driving sofware will lead to improvements in simulated driving. Whoopee! 🙄

“Navigant Research’s annual ratings of autonomous driving systems were just released. Once again,Tesla is rated as bottom of the pack.”

Great news for Tesla fans, thanks!
😆 😆 😆

Anyone paying attention knows that Navigant is always wrong — utterly clueless — when it comes to the EV market.

My problem with what Tesla is doing is that I don’t trust my ability to maintain situational awareness if I am not actively driving and then take over after I lose situational awareness.

Please, keep me from rear ending the car in front and the collision avoidance is great but I will keep overall control.

For me it is all or nothing. Either I am driving with assistance in emergency situations or I don’t even have a steering wheel.

I have the level 2 self driving features on a 2014 Infiniti. I zone out all the time in stop and go traffic, I cede responsibility for stopping to the car, and let me tell you its awesome! I can’t wait until I can have the car do everything else.

@Wil- Well, as driver, you get to choose just what Level of autonomy you wish to use, and when😊

If I could just get highway driving hands off that would make me happy. Right now though Tesla still has a ways to go on this without ending up at the butt end of some road rage accident. The car still drives fairly erratically in regards to speed, it really needs to smooth out all of it’s transitions and not do the speedup slow down dance.

Do you own, or have you driven a Tesla for more than 30 minutes, because that hasn’t been my experience😕

I have one.
Even with follow distance set to 1 (most aggressive), it drives like grandpa. People cut in front of me.
On a serene 2 hour drive, I had to take over twice or it would have crashed.

Yes, I like it.
No, I don’t trust it.
It’s great when traffic is moving.
Like Adam, I want highway driving hands off.

Back to this article, go check out Navigant’s staff backgrounds. They don’t know what they’re talking about.
I know Cruise AP too, total POS. Go SF, try to follow one and drive normally if you can.

In all fairness to Tesla, yes on the 1 setting it could be more aggressive but have you seen the aggressive drivers who cut everyone off nowadays? And that happens whether with autopilot on or driving the old fashioned way with two hands on the wheel! Not even Level 5 will protect a car in every situation. However when Level 5 accidents occur (and they will for whatever reason including those super aggressive drivers) will it legally be the car manufacturer or self-driving software manufacturer or the driver or the owner’s responsibility? I find the current autopilot iteration on my 2018 P100D much better than the other systems I’ve tried (2018 Mercedes SUV, 2017 Volvo Propilot). I can actually use it at any reasonable speeds and it captures more situations than the others. I couldn’t use the others without really feeling like a crash was ready to happen. And AP2.5 is definitely better than my AP1 (2016 X90D) so AP is improving over time.

“Source” of many numbers / charts in the video: ARK Invest.

Who is ARK Invest?

A company pumping Tesla stock with a price target of $4000.

That would make Tesla the most valuable company on the PLANET.

The same Tesla with a very thin liquidity cushion closing many stores…

Don’t worry, Tesla has found the perfect solution to the cash problem by unveiling the model Y today, hoping to rake up a few hundred of thousands of preorders.
The actual car is due next year, by then they hope they will have solved the miriad of problems they are still facing today, if not they can always unveil the pickup and get some more cash….

This is just FUD from the usual detractors. If Tesla really needed cash urgently, they’d just go to the capital markets. A March unveiling for Model Y had been stipulated like a year ago…

Model Y Reveal date was tweeted a year ago. Every single Tesla funding has been vastly oversubscribed.

“Every single Tesla funding has been vastly oversubscribed.”

We will see how Tesla does in a future recession (as you may remember from 2008-2009, car makers don’t do so well in down cycles…) and who is willing to lend them money then.

Tesla started selling its first car, the original Roadster, in 2008. Perhaps you have problems remembering Tesla’s history, Tftf; that would explain why you keep repeating the same mistake of short-selling TSLA over and over.

@Hans:

Dude, you’re playing a broken record. If you must come here only to post anti-Tesla pravduh B.S., then at least try to say something that hasn’t been repeated thousands of times.

Larger companies have more problems than smaller ones. Duh.

Not sure that would actually make them the most valuable company… But certainly way up there. However, *if* ARK’s assumption is correct that Tesla will be the first company with millions of fully autonomous vehicles on the road before anyone else, that valuation might very well be deserved. Whether they actually have a shot at that is hotly debated, of course…

Liquidity or closing physical stores has absolutely no relation to that. It doesn’t affect long-term prospects. (Unless the company actually went bankrupt of course — but no serious analyst sees a meaningful risk there.)

I’d say we are a very long way from knowing what company will be the first to develop a fully autonomous vehicle that’s more reliable than the average human driver. But at the moment, it’s Waymo that seems to be making significant advances, while Tesla is fiddling around with slight improvements to its Level 2+ autonomy.

What Tesla and other companies aiming for fully self-driving cars should be spending their resources on, is developing a system capable of handling Level 4 autonomy. They’ll never do that without lidar or phased-array radar. It’s like fiddling around with improving the sails on a sailing ship, when what is needed is reliable steam/ diesel power.

@tftf- do you have data that contradicts the data supplied by Tesla to ARK? If so please share….

No he doesn’t because tftf is a self-admitted Tesla shorter and therefore has an inherent conflict of interest on anything related to Tesla.

In fact as a shorter, he has a vested interest to come here and spread his serial anti-Tesla FUD because he thinks he can influence the stock price by doing that which shows you how delusional he is!

“No he doesn’t because tftf is a self-admitted Tesla shorter and therefore has an inherent conflict of interest on anything related to Tesla.”

LOL.

And ARK Invest with a price target of $4000 and managed funds full of Tesla shares hasn’t that conflict of interest?

Don’t know and don’t care. The point is that you never, ever tell the truth about Tesla or its cars. Your comments have zero informative value.

Entirely correct. The -2 down-votes must have come from Tftf and his fellow TSLA shorters.

This is an absolutely illiterate review of autonomous development. Currently Tesla L2 AP services are only mid-pack at best. Production L2 services from Volvo, Jaguar, Nissan and Tesla are all functionally just as good as each other. To suggest that Tesla’s current systems have so sort of advantage is just a clueless assertion. One area that Tesla does lead in with L2 systems is driver and passenger deaths that are directly attributable to L2. I would suggest the author get some real world information besides Musk blowing smoke out of his ass.

???

Disagree. Drive each of them yourselves and come to your own conclusions not a paper report. I tried the Volvo and Mercedes systems and the Tesla system was superior fundamentally in confidence in avoiding crashing into the car in front or not bouncing in the lane or leaving the lane and flying into some other vehicle. YMMV but I think most will get the same result on the latest AP2.5+. I used to be a LEAF lover as well (December 2011 new buy) but the battery treatment and residual values mean that I probably would be ahead financially if I had bought the Tesla S40 that I jealously badmouthed at that time.

Is “Leafover” another sock puppet for the serial Tesla bashers?

Tesla is more aggressive in enabling its semi-self-driving features in its cars. You can use AutoSteer on most roads, whereas (for example) GM’s Super Cruise won’t function anyplace except freeways.

In other words, Tesla’s suite of semi-autonomous features have broader capability than any other auto maker’s. (I think Waymo is more advanced, but they aren’t making cars.) So in that sense, Tesla is more advanced. On the other hand, there seems to be a consensus from reviewers that GM’s Super Cruise does a better job of smooth lane-keeping than Tesla’s AutoSteer, so if you choose to focus on that alone, then you might claim GM is “better” than Tesla.

(continued…)

(…continued)

It’s really subjective. Nobody except Waymo has advanced beyond Level 2 or 2+ autonomy, so IMHO these arguments amount to splitting hairs.

“One area that Tesla does lead in with L2 systems is driver and passenger deaths that are directly attributable to L2.”

It’s pretty clear that Tesla also leads, on a per-miles-driven basis, in saving drivers and passengers from serious accidents, injuries, and deaths. But of course, that doesn’t fit your anti-Tesla agenda, does it?

Your complaint there only means Tesla’s driver-assist features have not achieved absolute perfection in safety, so your Tesla bashing is just another case of “The perfect driving out the good.”

We will never get 100% safe self-driving cars. But we certainly can get self-driving cars which are safer than human-driven ones. Arguably, Tesla has already achieved that, altho of course Tesla bashers will never admit it.

All tesla drivers don’t use or did not purchase the software unlock option. What is the real percentage?

If anyone reads the comments before watching the video, don’t bother watching. It’s just a rehash of the years-old “billions of miles of data” theme. If you have a real interest in Autopilot and Self-driving, which are two vastly different things, ignore Ben Sullins and read this Forbes article by a someone who actually knows a thing or two about the topic (and also happens to own a Tesla):
https://www.forbes.com/sites/bradtempleton/2019/02/27/a-robocar-specialist-reviews-the-tesla-autopilot/#4c8e3444542a

Note, the article links to an update with some additional info.

The most pertinent part:

” It is not, however, anywhere close to a proper full self-drive offering. In fact, I would judge Autopilot to still be behind where Google/Waymo was in the early 2010s — and Waymo is still in 2019 driving mostly with safety drivers, after billions of dollars and work by some of the best minds in the business, with superior hardware. To be fair, Waymo could have produced a highway capable car by this time had that been their goal, but it still would have been a serious effort. The real problem of self-driving is deceiving. You can get 99.9% of the way there with a serious but tractable effort. That last tiny bit takes probably 100 times that effort. This has confused many teams into thinking they are “almost there” when they are not.”