I doubt there's ever going to be a big bang moment where self driving just works everywhere. Instead it's going to be a slow rollout of self driving taxi's in increasingly different environments.
And similarly: the transition will also go in the other direction. We'll start making roads and navigation easier for self-driving cars and prioritize the destinations we care about the most. At the tail end of the transition, there will still be areas with ridiculous intersections and confusing rules that only humans can do. We just won't care about them as much by then because everything we do care about is reachable by self-driving cars.
We didn't redesign roads and navigation for taxis or rideshares, and we're not going to do that for self-driving cars either.
We already do this stuff, and we won't need to rebuild 99% of roads. It's the "less than 1%" that are already tricky for normal humans that might need a rework. Or self-driving cars will just take suboptimal routes, if those roads / intersections aren't the only ways to important destinations.
Where I live in Seattle, they have rolled out significantly more 5-30 minute loading only zones on former street parking to deal with the uptick in rideshares and food/parcel deliveries, because the alternative is a bunch of illegal double parking.
LLMs are already better at this than humans:
https://x.com/petergyang/status/1707169696049668472
I think it's actually going to be the other way around. We'll build infrastructure exclusive to AVs where the rules are too complicated for humans, but allow AV traffic to move more efficiently. For example, an AV shouldn't need to stop and wait at a red light if there's no traffic to wait for.
At 2 trips per day for 300M Americans over 7 days, that would put the rideshare takeover at ~4.2Bn. If we extrapolated based on the referenced graph and exponential growth, that would put the takeover at 2029 :)
Its safe to assume that the limiting factors will soon become sourcing of components of the perception and control stacks.
This intersection is a local favorite, a backed-up lane off the highway onto a half-rotary that splits off into multiple directions: https://www.google.com/maps/@42.3560154,-71.1859078,17.52z?e...
What's hard for a robot is dealing with crazy unpredictable people, and while Boston has its share of those, I think SF is worse.
Way too much of this is developed in places that... don't have weather.
It's been standard in cars for a long time. It's a really pretty simple system that ensures that your wheels aren't just completely stopped while you're braking hard on slick services. When the wheel's are completely stopped there's less friction than when they're still rolling near the speed of the road. Also when your wheels are locked steering is nearly impossible, but you do have directional control while they're still rolling.
Some of the systems are more advanced than others, but the basic version just compares the speed of the wheels to each other.
Driving on snow is just an exercise in modeling friction in a computer and which way the car will go given available sensor input. Self driving cars don't innately have a concept of friction the same way a human with feet does, and they're able to drive on static asphalt, and also snow, with some training. Human drivers should practice driving in the snow in a parking lot to understand how the car slips and slides and grips operates under those conditions before taking to the road. (It's also fun!)
I'm sure it'll take a lo of doing to winterize the sensor packages and for the software to work well enough to be reliable when there's just snow on the ground that hasn't been plowed recently, nevermind when it's actively snowing. But personally I think it's a when and not if as to whether or not self driving taxis will ever hit New England. (No guess as to a specific timeframe though, lol.)
I'm not sure what you consider "real" weather, but those are all places I think qualify by any reasonable definition.
L4 in an urban environment is weirdly feasible by comparison, even though we still don't have examples of extremely good L2/L3 systems yet (there's far more complexity by default compared to highways: lights, pedestrians, complex lane decisions, more visual noise, higher average relative speeds) we already have these L4 Taxis being rolled out. The reason is that if a driver doesn't take over it's much easier to find a "safe" place to stop. Apart from getting stuck in the middle of a junction, you can feasibly park up anywhere that has speeds <=40mph.
The requirements for a self-driving vehicle on highways is very different from urban and and very different from sub-urban. Multiply that with how different countries and cities have very different looking roads, e.g. narrow windy roads in London vs wide straight roads in Phoenix.
I agree it's not going to be an location-by-location rollout (other than for regulatory reasons). But I do think Tesla will find that their product "just works" in some places, and doesn't work at all in others - at least for a time. I don't think Tesla would sit on their hands and say "it's not ready yet" until they'd cracked everything. If it's safe to take a nap on Highways and not safe in an urban area then I think they'll just start selling that.
Here's a good and relatively recent (3 months old) video about how its current capacities:
https://www.youtube.com/watch?v=6qY51q2Zifc
If you want to see how the latest beta drives, it's version 12.5.6: https://www.youtube.com/results?search_query=tesla+12.5.6 .
It's still far from ready to release, but it is also improving very quickly - far from solely hype. They have the most impressive non-geolocked self-driving system as far as I'm aware.
Especially with Tesla's Ceo soon to be a member of the government.
Just like cars companies made jaywalking illegal or bought public transport companies to close them.
Tesla FSD already works everywhere, even on unpaved roads. It just doesn’t work as well as Waymo.
Waymo works very well, just not in as many places as Tesla.
You might bet on Waymo because they have a fully working product already, but I’m betting on Tesla because of the vast amount of training data they are collecting. There’s a bitter lesson here.
And those demos are VERY old at this point.
I own a Tesla, though I don't own FSD, but this year, Tesla has given all cars a trial of FSD on two occasions. It works remarkably well. I backed out of my driveway, then enabled FSD and it drove all the way across Portland to a friend's place with zero intervention. It was about a 15 mile, 30 minute drive.
It navigated neighborhood roads without markings and tons of cars parked on the curb. It got onto the freeway and navigated, including changing lanes to overtake slow traffic. Once I got to their place, I was able to tell it to automatically parallel park on the curb.
As far as I'm concerned, Tesla has fulfilled their promise of full self driving. The "supervised" requirement is basically just being used as a legal loophole to avoid liability if it fails.
"If it fails" - so it is supervised for a reason then. It makes sense because FSD has an intervention rate in the low double digits according to community trackers like https://teslafsdtracker.com.
But it isn’t obvious to me that better sensors outperform better data.
Shipping working product should count!
The diversity of geography may be critical, though. You can only drive the Embarcadero so many times before your loss bottoms out.
Like cable/fiber, once they have good models of the business and what it costs to roll out, they have the freedom to accelerate and do regions in parallel. If the business works, I would expect them to scale the pace of rollout.
Now 150k trips per week (things are moving fast)
More data does not necessarily mean better data. You can collect many more individual driver experiences, but if they do not have sufficient resolution in the necessary dimensions, they may never provide “better data.” Similarly, even if the magic data is hidden somewhere in there, if the model cannot practically extract the insight because of their sizes/disorganization vs the computational/storage capacity, this too would mean they are not better data.
Of course you can make the argument that some of the sensors are unnecessary, but when one fleet has had millions of vehicles for years and isn’t working, and one started with dozens, has recently grown to one thousand vehicles, and is working, the evidence is not in support of the argument.
I would not bet on Tesla's FSD other than on highways. Same as many of the Tesla FSD owners I know.
Maybe the end-to-end NN version is better, though. I haven't been able to try it (hw3).
Waymo works any time, except where it doesn’t.
Tesla works any where, except when it doesn’t.
I might argue that every traffic light is sort of a where too. Mystery meat yellow light handling is scarily bad.
In contrast, Tesla is using dumb cameras and just dumping boatloads of data into their model. It’s a more general solution. Maybe the reference doesn’t fit perfectly - the model architecture is likely similar under the hood - but there’s some analogy there.
Just saying they have better results because of mapping and lidar is incredibly reductive. They have an extremely sophisticated AI/ML stack and simulators.
Start here: https://www.youtube.com/watch?v=s_wGhKBjH_U&t=2135s
Always remember that Full Self-Driving (Supervised) (also known as Autosteer on City Streets) does not make Model Y autonomous and requires a fully attentive driver who is ready to take immediate action at all times.They keep pushing this point. And they do appear to be collecting an absolute firehose of data from the millions of vehicles they have on the road. By comparison, Waymo collects a lot less data from many fewer vehicles.
Which leads to some tough questions about Tesla's tech. If they have (conservatively) 10x the training data that Waymo has, why can't their product perform as well as Waymo? Do they need 100x? 1,000x? 10,000x?
Assuming they were at parity with Waymo today, this would suggest that their AI is only at best 10% as effective as Waymo's, and possibly more like 1% or 0.1% or whatever. But since they can't achieve parity, it's not even possible to bound it.
It's entirely possible that their current stack cannot solve the problem of autonomous driving any more than the expert systems of the 60s could do speech translation.
I haven't heard a compelling argument as to why a system that is at best 10% as effective would ever be expected to be the leader.
Also, Tesla collects data from its fleet, but that data’s fidelity is likely quite limited compared to other companies, because of bandwidth if nothing else. Waymo can easily store every lidar point cloud of every frame of driving.
I hope for the best for Tesla, but they are many years behind Waymo. The world definitely needs a second working self-driving system! Right now comparing Tesla and Waymo is nonsensical. Once you can sit in the backseat of a Tesla while it drives there might be some worthy comparisons to be made.
I'll take that bet...
I predict the Chinese, in a decade, will have the first FSD
Tesla is miles behind
I'll take that bet any day. China and innovation don't go hand in hand.
Tesla works nowhere as a fully autonomous vehicle.
I also don't see any evidence that Waymo can't work anywhere. They recently expanded to Austin, and it seems that it immediately drives better than FSD.