In case your invitation to the World Artificial Intelligence Conference in Shanghai got lost in the mail, don’t worry because Elon Musk couldn’t make it in person either, instead delivering a video that answered a number of questions posed to him regarding AI and, significantly, Tesla’s Autopilot semi-autonomous system. In this video, Musk claimed that Tesla is “very close to level five autonomy,” and that there are “no fundamental challenges,” just “many small problems.” I’m not so sure I agree with Elon’s assessment here, mostly because the hurdles to true Level 5 autonomy are those “many small problems.”
Here’s what Elon said in response to the question “How confident are you that level five autonomy will eventually be with us? And when do you think we will reach full level five autonomy?”
“I’m extremely confident that level five or essentially complete autonomy will happen, and I think it will happen very quickly.
I think at Tesla, I feel like we are very close to level five autonomy. I think I remain confident that we will have the basic functionality for level five autonomy complete this year. So, yes, I think there are no fundamental challenges remaining for level five autonomy. There are many small problems. And then there’s the challenge of solving all those small problems and then putting the whole system together, and just keep addressing the long tail of problems.”
Just to remind everyone, Level 5 autonomy is completely self-driving. That is zero interaction is required from the driver past informing the vehicle where you wish to go, and Level 5 works without any restrictions, anywhere.
Level 5 autonomy has to work in our glorious and bafflingly chaotic real world, and Musk acknowledges this with his next statements:
“So you’ll find that you’re able to handle the vast majority of situations. But then there will be something very odd. And then you have to have the system figure out a train to deal with these very odd situations. This is why you need a kind of a real world situation. Nothing is more complex and weird than the real world.”
Now, here Elon and I are in total agreement: the real world is complex and weird. That complexity and weirdness are part of those “many small problems” Elon mentioned before, and it’s all part of why I do not think Tesla is even close to achieving Level 5 autonomy, certainly not by the end of this year.
Couple that with this statement from Elon:
“But I’m absolutely confident that this can be accomplished with the hardware that is in Tesla today.”
Okay, even if we decide to accept that Tesla doesn’t need lidar systems in their cars (I’m not convinced of that, but, whatever) the current hardware on Teslas is inadequate for one of those many small problems with Level 5 autonomy: dirt.
Full autonomy means that the car has to handle every situation, since a Level 5 car may not even have a steering wheel or any driver controls, and a very common situation is driving through weather, something we have here on Earth, the target market for these cars, and it’s likely that mud or dirt or grime may splash onto a Tesla’s camera lenses or radar emitter window, disabling or severely impairing that particular sensor.
On a highway at 80 or so mph, this can be a very big deal, and no Tesla on the market has hardware to deal with this very small-sounding but potentially life-or-death problem.
Ever driven through the South in the summer and had to scrape off entire swarms of bugs from your windshield? Just a few juicy flies could obscure the tiny lens area of a camera, for example.
It’s hard to imagine any issue more common than a car getting a little bit dirty, but the truth is a Tesla with autonomous software and current hardware could be crippled by one muddy puddle or a good-sized swarm of mosquitoes.
Again, this is just one of those small problems Elon mentioned. There’s also the wildly varying behaviors of how humans cross streets, or gesture their intentions to people driving cars, or confusing construction sites, or any number of other “small” challenges that are actually a very normal part of our daily driving lives here in the messy real world.
Dismissing the real world situations as just “many small problems” while insisting that Tesla is ready for Level 5 this year makes me think that Elon either doesn’t really appreciate the scale of full Level 5 or is deliberately downplaying the challenges.
Tesla hasn’t even mastered Level 2 semi-autonomy yet, despite pre-selling their “Full Self-Driving” system that simply doesn’t exist yet. We’re still seeing wrecks where Tesla’s Level 2 system gets confused and fails and have yet to see any evidence of anything approaching Level 5.
Tesla’s low-speed, limited use Enhanced Smart Summon feature was rolled out as a beta test with pretty iffy results, which is a far, far cry from Level 5 driving as well.
The Center for Auto Safety is skeptical as well, issuing this statement regarding Musk’s remarks:
“Tesla can make as many unverifiable claims as they want, but as we’ve continued to see they are still working hard on getting level two right. Let’s not forget, the other day a level 2 Tesla slammed into a stopped police car in Connecticut, in the middle of a well lined road, on a clear day. To avoid this charade, and consumer and investor deception, NHTSA could have written performance standards for features like automatic emergency braking or forward collision warning, or even could have required driver monitoring systems for vehicles with ADAS features to avoid automation complacency. But of course that hasn’t happened. Instead, expect NHTSA to respond by making a feckless statement about potential after-the-fact safety enforcement that is as separated from reality as Musk’s statement about Tesla’s fully driverless capability arriving next year.”
I think the problem here lies in the fact that Level 5 seems to be seen as some sort of hard-and-fast marker point, when really it’s more like a gradient, a goal that we can keep striving for, but perhaps asymptotically never quite achieve.
That’s not to say I don’t think we can get very, very close, but I think true, complete Level 5 driving is vastly more complex than Elon or many other developers seem to think, as it incorporates far more nuance and cognition that is wrapped up in all sorts of human communication and culture, situations far more advanced and subtle than just reading road signs.
When Elon says they have the “basic functionality” that’s fine, but it’s also just a tiny part of what Level 5 autonomy will need to be.
I think the future will most likely be one of mostly Level 4 autonomy; that is, cars that effectively drive themselves, but in areas that have been designed, on an infrastructure level, to accomodate AVs.
This may be lanes on a highway or whole sections of cities, but the only way to really make autonomous vehicles safe and reliable is to meet them partway. Transportation planners will need to figure out ways to limit the set of chaos that is the world to a more controlled subset, one that AVs can reliably handle, and those are the places AVs will exist and function.
This is a far more achievable goal than full Level 5, and is something that we may realistically hit in short-ish timelines, but it will take cooperation from carmakers and the organizations that run the areas they’ll be expected to roam.
I’m pretty comfortable saying that, no, Tesla will not have full Level 5 autonomy solved by the end of the year, especially not with current hardware on their fleet of cars. I do not think we will see a software solution to L5 downloaded to Teslas any time soon.
I don’t really understand why Elon is pushing this narrative, either. Tesla has been saying they’re just about to release “Full Self-Driving” for years now, and they haven’t.
Elon’s remarks both suggest he’s aware of the scale of the problem, yet he seems to trivialize the issues or ignore them, anyway. While I think it’s possible that enough of the issues for Level 5 can be engineered away to be viable, we’re not really close yet.
You said it yourself, Elon: the world is complex and weird. You need to respect that, and be honest about the challenges of truly full autonomy.