Will Tesla delay their robotaxi at the request of Austin representatives?
“As members of the Austin delegation in the Texas Senate and Texas House of Representatives, we are formally requesting that Tesla delay autonomous robotaxi operations until the new law takes effect on September 1, 2025. We believe this is in the best interest of both public safety and building public trust in Tesla’s operations,” the letter read. (Quote from article; Tesla says the current target date is this weekend, June 22, 2025)
The new Texas law they reference raises the bar, requiring prior authorization required from Texas DMV to remove the driver. It also changes the responsible party for traffic violations from the computer (which is nonsensical, but is current Texas law) to the owner/authorization holder. Other requirements as well more in line with regulations we have seen in other states that have become sophisticated about their AV safety approach. This situation marks the beginning of the end of the current Wild West situation for AV accountability in Texas. But it doesn't go into effect until September.
It is difficult to believe Tesla will wait just because of this request, unless they want to delay for technical reasons anyway. They might just ignore it. But they could also reply that they already voluntarily comply (if true) with all the technical requirements and just need to do the paperwork approval cycle. We'll see how this turns out.
Remote operations and control centers remain a mystery.
Everyone has one, but no one talks about it—except maybe in passing during an article or a rare interview. (efficient) scalability is impossible without optimizing the operator-to-AV ratio. I’d really appreciate more insight into how AV companies are approaching remote operations. I also don't expect this transparency from Tesla.
Phil, you mentioned the possibility of having more than one mobile operator connected simultaneously. However, switching between IP connections would need to be done in a fraction of a second, and as far as I know, this cannot be achieved with conventional mobile technology (i.e buying a more or less off-the-shelf module from an OEM with more or less standard software package).
To take an example, if you are at home with a wifi connection on an Android phone ( I have not tested this with an iPhone, but I very much doubt that such issues are a huge focus at Apple), and take a walk, it will generally take a couple of minutes before the phone switches over to the mobile network. This is because the phone will attempt to reconnect to the previous connection, even if the field strength of the old connection is very weak, in the hope that the problem is just a temporary radio shadow.
And of course, in many locations mobile operators will share the towers, so if one connection is lost, other connections using that tower will probably also be lost.
While it will of course be possible, at least in theory, to send two parallel data streams continuously, (given that they use different mobile towers) there still will be the difficulty of selecting which of these data streams is valid and which is not. I doubt that Tesla has the competence to build this in a reliable manner - or at all.
Too much speculation; it makes more sense to believe in what makes more sense. For example we know Tesla will be using a basic version of the Y instead of the paraded Cybercab which aren't even in production. Also, the Y is less conspicuous than the Cybercab at a time of intense negative perception of Musk and, by extension, of Tesla.
Today's version of FSD is clearly not suitable as named, despite continuous improvements over the years. Musk's aspirational Full Self Driving "by next year" is a bit more complex than expected, but I won't be the first to throw stones on this ambitious project which, I'm sure, will become reality some time in the not too distant future.
Consequently, Tesla will most likely hire dozens of remote operators in an anonymous building somewhere in Austin. Each Y cab will be remotely operated by 1 driver with full control of the vehicle (including steering) at least for the time being.
Since Tesla is operating, like Waymo, within a defined area, we can expect FSD to safely operate 70% of the time (?) and require the remote operators to be particularly vigilant of the remaining 30% possible unsafe mishaps.
However, within a small area crisscrossed by Y cabs, it shouldn't take long for AI to solution repetitive unsafe zones or quirks encountered by the remote operators. Within months we should expect the Y cabs to be safely operated 99.99% of the defined area. This would lead to a couple possible scenarios, 1. Would be to cut the number of remote operators to 1 for 5 or 10 cabs. Another more plausible scenario would be for Tesla to retain the 1 for 1 remote operation but request an extension of the drivable area 2-3x wider than Waymo's present limits.
The repercussions of Tesla's cab venture to its overall business is rather important. FSD's slow improvements stems in part because of the far too few FSD beta owners in light of the highly complex general driving requirements. Musk should have given it away to help speed up its development, but I think he really believed he had it solved years ago.
Today, by limiting FSD's area of operation, the AI can more quickly test solutions and resolve repetitive quirks. Increase the area of operation to multiple cities and their suburbs will warrant a level 4-5 approval and remival of remote operators.
Eventually that experience wil be transferred to the entire pool of Tesla owners who will, finally, have FSD.
Tesla's competitive advantage is AI, its data storage and years of experience transferring data to and from the millions of vehicles. I don't underestimate Chinese competitors' determination to match or surpass Tesla's FSD, but it will take long enough for Tesla to dominate the EU and North American markets for the foreseeable future.
Going without Lidar is the bet Musk has made for Tesla. Mobileye, the Lidar solution company is going out of business. Others, including Tesla, are testing various radar alternatives. In the end, they may or may not use visualization techs beyond camera only.
No, no... Mobileye shut down its Lidar business. But Tesla is one of Luminar's largest customer, which shows that Musk has not entirely abandoned the idea.
Will Tesla delay their robotaxi at the request of Austin representatives?
“As members of the Austin delegation in the Texas Senate and Texas House of Representatives, we are formally requesting that Tesla delay autonomous robotaxi operations until the new law takes effect on September 1, 2025. We believe this is in the best interest of both public safety and building public trust in Tesla’s operations,” the letter read. (Quote from article; Tesla says the current target date is this weekend, June 22, 2025)
The new Texas law they reference raises the bar, requiring prior authorization required from Texas DMV to remove the driver. It also changes the responsible party for traffic violations from the computer (which is nonsensical, but is current Texas law) to the owner/authorization holder. Other requirements as well more in line with regulations we have seen in other states that have become sophisticated about their AV safety approach. This situation marks the beginning of the end of the current Wild West situation for AV accountability in Texas. But it doesn't go into effect until September.
It is difficult to believe Tesla will wait just because of this request, unless they want to delay for technical reasons anyway. They might just ignore it. But they could also reply that they already voluntarily comply (if true) with all the technical requirements and just need to do the paperwork approval cycle. We'll see how this turns out.
Bill text here: https://capitol.texas.gov/tlodocs/89R/billtext/pdf/SB02807F.pdf#navpanes=0
The article has more details: https://www.kxan.com/news/local/austin/texas-lawmakers-send-letter-to-tesla-asking-to-delay-robotaxi-rollout-in-austin/
Remote operations and control centers remain a mystery.
Everyone has one, but no one talks about it—except maybe in passing during an article or a rare interview. (efficient) scalability is impossible without optimizing the operator-to-AV ratio. I’d really appreciate more insight into how AV companies are approaching remote operations. I also don't expect this transparency from Tesla.
Thank you for the article
Hi Phil, thanks for your take! Will be interesting to watch what happens this week. Hopefully no one is getting hurt.
Keep up the good work!
Phil, you mentioned the possibility of having more than one mobile operator connected simultaneously. However, switching between IP connections would need to be done in a fraction of a second, and as far as I know, this cannot be achieved with conventional mobile technology (i.e buying a more or less off-the-shelf module from an OEM with more or less standard software package).
To take an example, if you are at home with a wifi connection on an Android phone ( I have not tested this with an iPhone, but I very much doubt that such issues are a huge focus at Apple), and take a walk, it will generally take a couple of minutes before the phone switches over to the mobile network. This is because the phone will attempt to reconnect to the previous connection, even if the field strength of the old connection is very weak, in the hope that the problem is just a temporary radio shadow.
And of course, in many locations mobile operators will share the towers, so if one connection is lost, other connections using that tower will probably also be lost.
While it will of course be possible, at least in theory, to send two parallel data streams continuously, (given that they use different mobile towers) there still will be the difficulty of selecting which of these data streams is valid and which is not. I doubt that Tesla has the competence to build this in a reliable manner - or at all.
To be clear, it is one remote operator managing multiple vehicles, not the other way around.
The idea would be to reduce the operations cost by having fewer drivers than vehicles.
Too much speculation; it makes more sense to believe in what makes more sense. For example we know Tesla will be using a basic version of the Y instead of the paraded Cybercab which aren't even in production. Also, the Y is less conspicuous than the Cybercab at a time of intense negative perception of Musk and, by extension, of Tesla.
Today's version of FSD is clearly not suitable as named, despite continuous improvements over the years. Musk's aspirational Full Self Driving "by next year" is a bit more complex than expected, but I won't be the first to throw stones on this ambitious project which, I'm sure, will become reality some time in the not too distant future.
Consequently, Tesla will most likely hire dozens of remote operators in an anonymous building somewhere in Austin. Each Y cab will be remotely operated by 1 driver with full control of the vehicle (including steering) at least for the time being.
Since Tesla is operating, like Waymo, within a defined area, we can expect FSD to safely operate 70% of the time (?) and require the remote operators to be particularly vigilant of the remaining 30% possible unsafe mishaps.
However, within a small area crisscrossed by Y cabs, it shouldn't take long for AI to solution repetitive unsafe zones or quirks encountered by the remote operators. Within months we should expect the Y cabs to be safely operated 99.99% of the defined area. This would lead to a couple possible scenarios, 1. Would be to cut the number of remote operators to 1 for 5 or 10 cabs. Another more plausible scenario would be for Tesla to retain the 1 for 1 remote operation but request an extension of the drivable area 2-3x wider than Waymo's present limits.
The repercussions of Tesla's cab venture to its overall business is rather important. FSD's slow improvements stems in part because of the far too few FSD beta owners in light of the highly complex general driving requirements. Musk should have given it away to help speed up its development, but I think he really believed he had it solved years ago.
Today, by limiting FSD's area of operation, the AI can more quickly test solutions and resolve repetitive quirks. Increase the area of operation to multiple cities and their suburbs will warrant a level 4-5 approval and remival of remote operators.
Eventually that experience wil be transferred to the entire pool of Tesla owners who will, finally, have FSD.
Tesla's competitive advantage is AI, its data storage and years of experience transferring data to and from the millions of vehicles. I don't underestimate Chinese competitors' determination to match or surpass Tesla's FSD, but it will take long enough for Tesla to dominate the EU and North American markets for the foreseeable future.
It will never work reliably without lidar.
Going without Lidar is the bet Musk has made for Tesla. Mobileye, the Lidar solution company is going out of business. Others, including Tesla, are testing various radar alternatives. In the end, they may or may not use visualization techs beyond camera only.
I think you mean Luminar, but there are plenty of lidar manufacurers.
No, no... Mobileye shut down its Lidar business. But Tesla is one of Luminar's largest customer, which shows that Musk has not entirely abandoned the idea.
Phil,
I know that many people read your articles without commenting because they’re so good. I’m one of them. Just want you to know!
Jim, thanks for the kind words. This really made my day!