A more rational approach to “green.”
1 Like
Has anyone been in a robo taxi? Looks like a Chevy Bolt?
1 Like
SF is one of the most complex driving environments in the US. I think it’s even more challenging than Manhattan, but we don’t have snow. Driving in snow is another huge ball of wax.
Despite what Musk claimed year after year, Tesla self driving is nowhere near prime time. GM, the owner of Cruise, is light years ahead of Tesla.
1 Like
1 Like
The plain truth is EM wants to leave Tesla and Twitter to someone foolish to run them. His current interest/ focus is SpaceX.
Foolish = Easy (fun?) jobs are done. What left are tough (boring?) nuts to crack. EM wants out.
1 Like
Share holders are applying pressure on EM, but it is hard to remove him.
Twitter is his new private baby, he needs it to make it big like Google (social media) and Facebook (social media) !
He will squueze twitter and make it big and that is more than enough sales/advertising business for all his empire.
He has so many companies/activities
-
SpaceX,
-
Starlink,
-
Boring company for hyperloop and
-
Neuralink and OpenAI ( Neuralink Corp is developing a brain implant it hopes will help paralyzed people walk again and cure other neurological ailments.)
1 Like
This may not necessarily be true. Of course I am no expert in this so this is just a layman’s observation. What you can do depends a lot on what your system is capable of doing. Tesla obviously was overly confident on their current hardware implementation (or maybe they knew from the beginning that it was not sufficient but didn’t want to say it publicly). But now that’s in millions of cars, they are limited by what that system is capable of doing, which is now proven to be insufficient. That’s why they are proposing a new level of hardware or even bringing lidar back.
Other companies don’t have the same kind of restrictions, because they have not committed to a set of hardware like Tesla did. They can use processors that are faster, cameras that have higher resolution, etc. They might be lightyears ahead of Tesla on self-driving tech but it doesn’t mean that they are ahead in terms of a commercially-viable self-driving solution. Waymo’s business plan is not to equip car manufacturers but only to run a robo taxi business because their technology cost is too high.
2 Likes
Just different business strategy.
-
No LiDAR, no self-driving, lower manufacturing cost, can sell cheaper, capture large market share and profit. With a “captive” market, introduce LiDAR and self-driving. Market will buy because of familiarity and habit… 101 of branding.
-
Got LiDAR, can do self-driving, high cost of manufacturing, price of car too expensive, can’t sell much, slow operational efficiency improvements, no profit to fund R&D, bet for angel money, pray for component prices to decline and more subsidies.
When self-driving cars can go 10-15mph over the limit, see the CHP before he/she sees the car and slow down then I might consider one.
I did that last night - just about. Saw him in the last second. No ticket. At least 15 mph over the limit. He got on my tail for some, followed me past two speed limit changes as I drove into town figuring he knew I WAS speeding at one point and maybe I’ll do it again. He gave up and turned off. Like always. I’m the lead-foot whose last speeding ticket was in 1997. I’ll know I’m getting old, slipping and losing my edge when I get the next one.
Can self-driving cars be programmed to do that?
Well?
Can they 
My pessimism on Tesla FSD is not just their inadequate hardware, but also their whole approach. To this day they have zero miles of fully automatic driving on city streets. All their testing miles are done with humans in the loop.
This has huge implications. Software that is designed and tested to have humans as fail safe system, vs Cruise and Waymo that have to have redundancy designed in and tested on real roads. So to me Tesla’s millions of FSD miles is irrelevant. They are just testing their L2 system.
Another issue is their “boil the ocean” approach. FSD is tested everywhere by everyone all at once. Versus the limited geofenced areas approach pretty much by everybody else. Do we really expect Tesla can skip the limited area testing and leapfrog everyone by just flipping a switch? I don’t think so. Another thing that tells me how far behind Tesla is.
Actually, their system is ideal IMO.
By deploying fleet wide they have the ability to gather immense amount of unique data, which is essential in creating a robust AI system.
By having a driver behind the wheel that disengages when the system screws up, Tesla has essentially crowdsourced filterers of data so that only the relevant data gets downloaded and annotated. And as the driving models improve, the data flow will decrease as disengagements decrease. But Tesla will sell it to more people, and they will use it more often, so the data stream can sustain for some time.
Waymo / Cruise having nothing like this at the same scale, and they will end up with multiple orders of magnitude of less “useful” data.
Is that data necessary? As a data scientist, my intuition is “most likely” based on all the problems I’ve worked on. Self driving is quite possible the most complicated challenge of statistical learning so far, and we think very limited data (especially geographically) will be sufficient?
And the fact is so far, Lidar has prove unnecessary for Tesla, their computer vision is “good enough” for now. Note I think in the future it could be necessary to add Lidar to reduce error rate, but it is not impeding progress right now.
Limited compute power is a definite consideration I think. Tesla doesn’t even have all planning / prediction modules as NNets yet. Once they do, I think improvements should be significant.
That said, if they want close to Waymo performance, I think they are going to have to augment map data better.
All in all, I think Tesla is certainly behind Waymo in tech, but they have a few years to improve software before Waymo is even close to scaling. That’s really not unreasonable.
And the longer it takes companies to scale to cash-flow neutral levels, the better positioned Tesla is.
From a business perspective, they are taking the right approach.
1 Like
Being 80% or even 90% is not good enough, certainly not good enough for Tesla to assume responsibilities in case any accidents happening.
I just don’t think Tesla’s approach with ever get them to the 99.9% that can be described as fully self driving. For an L2 system 80% is good enough though. I suspect that’s what Tesla is after all along, despite what Elon said publicly.
1 Like
Why not?
BTW, BYD revenue down 23% QoQ, Tesla was down 4%. Ouch.
Because of the things I wrote earlier. The stuff they are testing now has a human always in the loop. So you can say that they have zero autodriving miles under their belt. I don’t think it’s easy to “port” these L2 test miles to L4 and above.
There is no fail safe redundancy system built into their fleet right now, as it would be cost prohibitive to do so. What they are testing today has little resemblances to the final system they need to ship in order to achieve L4.
Versus Cruise now already can drive 24/7 in all of San Francisco, arguable the most challenging driving conditions west of Mississippi. I don’t see why cruise can’t roll their service to all of California. There is no technical reason why not, only economic ones. And Tesla hasn’t even solved the technical problem of self driving.
1 Like
Also Tesla removed the USS in their parking system. So Teslas can’t even safely park themselves in the owners’ garage much less driving themselves on public roads.
EM has the attention span of a cocker spaniel
Because of the things I wrote earlier. The stuff they are testing now has a human always in the loop. So you can say that they have zero autodriving miles under their belt. I don’t think it’s easy to “port” these L2 test miles to L4 and above.
Why not? What exactly does the system do differently if it’s L2 vs L4?
How do you think Waymo and Cruise collected the majority of their miles for deploying L4? They had humans sitting in the drivers seat.
Cruise and Waymo didn’t transition to L4 while their system was still having frequent disengagements. They esentially were at L2 until their systems were reliable enough to try L4 without having too many issues.
Really cool interview with Rivian CEO. RJ came across as really authentic and a great guy.
1 Like