FSD rewrite will go out on Oct 20 to limited beta

[S4WRXTTCS]

Interesting, so you don't think Tesla will ever be able to perform as well as Waymo in Chandler? Assuming you've watched the Waymo videos

Chandler Arizona doesn't offer any profitability for either Waymo or Tesla.

I fail to see how it benefits Tesla to put any effort whatsoever in getting L4 approval in same geofenced grid as Waymo currently runs their money bleeding autonomous ride-hailing service.

Tesla has to achieve L4 driving in populated areas that will have challenging regulatory hurdles as that's where their sales are.

To make matters worse for Tesla the SAE will have guidelines for safety, and I don't believe the HW3 Sensor Suite will meet the guidelines. The populated areas will simply adopt the guidelines.

 

[jeremymc7]

Interesting, so you don't think Tesla will ever be able to perform as well as Waymo in Chandler? Assuming you've watched the Waymo videos

I've watched dozens of Waymo videos. The cars are much more expensive, the sensor suite is much more broad/deep, they themselves have plenty of disengagements that often don't make sense, they block out loads of areas that seem to be more about where the car just can't function right.

I've also seem plenty where the car gets stuck, road service gets called out to move fix it, car starts going, service gets canceled, cycle repeats for 10-15 until service just decides to finish the drive for you with a person in the driver seat.

 

[jeremymc7]

The new Polestar concept promised to be entering production (with unknown changes) has LiDAR as part of their sensor suite. Radar is also included along with the HQ cameras, etc.

There is also a full LED rearview mirror (not switchable like GM) and side cameras (not mirrors) with displays in the doors (ala Audi Europe); good luck with those in the US with current outdated legislation.

For what it's worth.

Pricing is expected to be in Model S area.

 

[S4WRXTTCS]

On a limited access highway what kind of dangers do you think would be posed by it going to 85? It certainly seems the cameras are good enough for that.

It's really about conflicts of interest when the car is in L3 mode due liability issues.

Lawyers -> To reduce the chances of accidents, and severity lets limit the speed to the speed limit at all times
Owner -> How is that goes to work? Even truckers will pass me

Lawyers -> Lets limit the handoff time to the bare minimum
Owner -> That barely gives me time to reestablish situational awareness

Lawyers -> If the driver doesn't respond during the handoff time lets just have it stop in the road.
Owner -> well that sucks, but okay.

Lawyers -> To minimize the chances of the driver falling asleep lets use driver monitoring, and only support it when it can clearly see the drivers eyes
Owner -> But, I don't have the driver facing camera

Then there is Tesla themselves as they've expressed no interest whatsoever in L3. I can totally see them doing a "me too" thing when competitors have a traffic jam L3 system, and Tesla will simply match theirs. Tesla has a far superior multimedia offering so they have a leg up on that.

I don't think we'll see any freeway speed capable L3 system from any manufacture. Even Tesla won't touch it.

 

[Magellan55]

So how does the FSD beta and HW3 handle highway onramp/merging? This still sucks for me on HW2.5. My car never sees the one in the lane I'm trying to merge into, or sees a car two lanes over and thinks it's in the merge lane so doesn't merge. Either way it's a borked merge and I have to take over. No improvements in 3 years for me. And I suspect it's the sensors, not the CPU, as the cameras can't detect range in a merge situation. So I'm really interested to see if the FSD SW and HW3 have fixed this or not. Until it does, I don't see Tesla advancing past L2.

 

[jeremymc7]

Pretty sure that HW3 is only a computer change since HW2.5. They both have the same cameras, radar, etc.

I believe the reason that HW2.5 can't handle the visualizations that HW3 can do is down to processing and little else.

HW2 though is a much different story.

 

[MP3Mike]

HW2.5 can't run the Tesla Vision software. At least not the current version.

 

[diplomat33]

What I find most interesting is that prior to fsd beta, many of us were saying the sensor suite is impossible or pointing out some hardware limitation, but now with V9, we're not talking about hardware inadequacy anymore. Now we're pointing out disengagements or dumb decisions that have nothing to do with the sensors but rather the software or perception not seeing or acting on something that is obviously there.

There was never any doubt that the current sensors and hardware could do some self-driving like what we are seeing in the FSD Beta V9 videos. But there is a big difference between "doing self-driving" and "doing self-driving with a safety greater than humans and with no human supervision". Remember that Tesla wants a safety disengagement rate of 1-2M miles per disengagement. We are far from that goal. Can the current sensors achieve that goal? We don't know yet.

Also, could additional sensors help the car make better decisions? For example, We've seen plenty of instances where FSD Beta creeps forward a lot or waits a lot before making a turn, or has to veer a bit into the incoming lane to peek around a parked car. Sure, it works, especially with a driver that can tap the accelerator to help the car know when to go. But, could additional sensors help the car do better? For example, could a bumper camera help the car see around corners and cross traffic sooner where it does not need to creep forward as much and can make the maneuvers with even higher reliability? Probably. If more sensors could help achieve higher safety, where you can remove driver supervision and trust the car to drive with high confidence, that would be a good thing IMO.

I've watched dozens of Waymo videos. The cars are much more expensive, the sensor suite is much more broad/deep, they themselves have plenty of disengagements that often don't make sense, they block out loads of areas that seem to be more about where the car just can't function right.

I've also seem plenty where the car gets stuck, road service gets called out to move fix it, car starts going, service gets canceled, cycle repeats for 10-15 until service just decides to finish the drive for you with a person in the driver seat.

JJ Ricks did over 1000 miles, all documented in 78 videos and those issues you mention only happened a few times. It does not happen as often as you seem to think.

JJRicks.com

Welcome to the JJRicks Studios Video Archive. Here you'll find the full record of Waymo self-driving car videos I've produced-along with easy ways to view, sort, and search through every notable event that's ever happened during one of my recorded rides. Enjoy! Contents: Videos - List of all my

www.jjricks.com

Of course 1000 miles is still a small sample. We need to look at a bigger sample to get a more accurate statistic. In their CA DMV report, Waymo reported only 21 disengagements in over 600,000 miles last year. That's an average disengagement rate of 1 per 30,000 miles. So overall Waymo has very few disengagements.

 

[drtimhill]

It looks like @powertoold was a bit optimistic in his prediction last fall. Does it feel like we are closer to 150k miles between accidents?

I dont really think any statistics from the beta drivers would mean much, since they are instructed to be very alert and disengage if in doubt. I doubt the attention level will be as high for a wider audience.

 

[MP3Mike]

JJ Ricks did over 1000 miles, all documented in 78 videos and those issues you mention only happened a few times. It does not happen as often as you seem to think.

JJRicks.com

Welcome to the JJRicks Studios Video Archive. Here you'll find the full record of Waymo self-driving car videos I've produced-along with easy ways to view, sort, and search through every notable event that's ever happened during one of my recorded rides. Enjoy! Contents: Videos - List of all my

www.jjricks.com

Of course 1000 miles is still a small sample. We need to look at a bigger sample to get a more accurate statistic. In their CA DMV report, Waymo reported only 21 disengagements in over 600,000 miles last year. That's an average disengagement rate of 1 per 30,000 miles. So overall Waymo has very few disengagements.

And in that ~1000 miles he had what looks to be at least 9 disengagements. So using his sample we get 1 disengagement per 111 miles.

It seems like the driving they have done in California isn't really representative of the driving when they are running an actual Robotaxi service...

Where is the Waymo data on disengagements from their Arizona operations?

 

[jeremymc7]

JJ Ricks did over 1000 miles, all documented in 78 videos and those issues you mention only happened a few times. It does not happen as often as you seem to think.

Yeah, I've watched many of his videos as well.

Keeping in mind (as I'm sure your aware) again that Waymo also geofences, and as I believe Tesla (currently) doesn't. Often times just because the car has problems in to areas. So the sample isn't quite fair, or at least after a geofence goes up, as compared to prior. Once you have a know issue in certain areas and you lock the car out known problem areas that greatly reduces amounts of disengagements.

Sure there are also geofences because it's outside of a service area or it was requested to be excluded etc. But take a look at where all of it is, much of it doesn't make sense.

And of course as before Waymo also has way more sensors, in places with some better visibility, and presumably more detail, read more expensive.

Not knocking AP/FSD AA I bought in on one of my cars and use it dally.

 

[Cosmacelf]

This is a good video of the current 9.1 FSD beta. The author does not waste your time, every bit of it is informative. He shows everything, warts and all. Personally, I would take this version as shown in a heartbeat. Good enough for me. Not level 5, but close enough that the car could get me through my drives as long as I paid attention.

 

[powertoold]

Can the current sensors achieve that goal? We don't know yet.

I think it can. The real problem is whether or not hw3 cpu is enough.

 

[jeremymc7]

This is a good video of the current 9.1 FSD beta.

Actually just finished watching it before you posted this. HA!

 

[diplomat33]

And in that ~1000 miles he had what looks to be at least 9 disengagements. So using his sample we get 1 disengagement per 111 miles.

It seems like the driving they have done in California isn't really representative of the driving when they are running an actual Robotaxi service...

Well, with a driverless ride, there is no disengagement since there is no safety driver. There can be remote assistance but they don't necessarily disengage the autonomous driving. So I am not sure we can accurately compare "interventions" with a driverless ride to disengagements with safety driver testing.

Also, 4 of the CA disengagements were on the highway which indicates highway driving. The Chandler robotaxi service does not do highway driving. So yeah, the CA testing is not completely representative of the robotaxi service in Chandler.

Also, 1000 miles is statistically too small to get an accurate disengagement rate. You need a bigger sample to get a more accurate statistic.

The stat of 1 per 30,000 miles is an average. If you break down the CA disengagement report, there were some cars with a disengagement rate of only 1 per 1,688 miles. There were also some cars with a disengagement of 1 per 9,000 miles. And some cars that did over 10,000 miles with zero disengagements.

Here is the full table if you are interested:

VIN	Disengagements	Miles	Miles Per Disengagements
2C4RC1K74HR805516	3	5064.2	1688.1
2C4RC1K76HR534698	1	2944.6	2944.6
2C4RC1K75HR695706	1	3033.3	3033.3
2C4RC1K77HR805512	2	7609.7	3804.9
2C4RC1K7XHR534705	1	4238.5	4238.5
2C4RC1K7XHR805455	2	10591.8	5295.9
2C4RC1K71HR695699	1	6100.3	6100.3
2C4RC1K75HR797345	1	6463.7	6463.7
2C4RC1K76HR797404	2	12928.2	6464.1
2C4RC1K79HR833859	1	9243.3	9243.3
2C4RC1K70HR833796	1	9941	9941.0
2C4RC1K79HR822215	1	10584.3	10584.3
2C4RC1K73HR833842	1	11822.8	11822.8
2C4RC1K7XHR797325	1	12252.2	12252.2
2C4RC1K70HR827271	1	13079.3	13079.3
2C4RC1K78HR797324	1	17126.1	17126.1
2C4RC1K79HR797381	0	17269.9	N/A
2C4RC1K74HR827239	0	16757	N/A
2C4RC1K75HR797362	0	16047.2	N/A
2C4RC1K74HR833879	0	15439.3	N/A
2C4RC1K7XHR822188	0	14947.6	N/A
2C4RC1K76HR822205	0	14522.1	N/A
2C4RC1K78HR833772	0	13932	N/A
2C4RC1K74HR805497	0	13717.8	N/A
2C4RC1K70HR797379	0	13227.5	N/A
2C4RC1K78HR797355	0	12102.7	N/A
2C4RC1K73HR797358	0	11796.6	N/A
2C4RC1K75HR812331	0	11702.2	N/A
2C4RC1K77HR833827	0	10997.1	N/A
2C4RC1K73HR797389	0	10376.9	N/A
2C4RC1K7XHR827262	0	10302.5	N/A
2C4RC1K78HR797405	0	10142.5	N/A
2C4RC1K73HR797330	0	10014.5	N/A
2C4RC1K71HR797343	0	9965	N/A
2C4RC1K74HR805466	0	9477.8	N/A
2C4RC1K73HR833758	0	9163.7	N/A
2C4RC1K73HR534660	0	8918.9	N/A
2C4RC1K7XHR797308	0	8589.6	N/A
2C4RC1K78HR805518	0	8490.2	N/A
2C4RC1K78HR833867	0	8384.3	N/A
2C4RC1K78HR833769	0	7796.2	N/A
2C4RC1K72HR827238	0	7432.3	N/A
2C4RC1K7XHR827231	0	7010.2	N/A
2C4RC1K79HR824255	0	6754	N/A
2C4RC1K7XHR833787	0	6690.2	N/A
2C4RC1K72HR534665	0	6509.4	N/A
2C4RC1K72HR797318	0	6488.5	N/A
2C4RC1K72HR836179	0	6300.8	N/A
2C4RC1K70HR805478	0	6277	N/A
2C4RC1K70HR534681	0	5598.4	N/A
2C4RC1K73HR534691	0	5388.4	N/A
2C4RC1K7XHR805486	0	5388.3	N/A
2C4RC1K76HR534653	0	5201.4	N/A
2C4RC1K76HR833835	0	5167	N/A
2C4RC1K76HR695696	0	5088.5	N/A
2C4RC1K72HR534696	0	4438.5	N/A
2C4RC1K73HR695705	0	4026.9	N/A
2C4RC1K70HR833877	0	3980.3	N/A
2C4RC1K73HR797361	0	3974.3	N/A
2C4RC1K74HR695700	0	3774.3	N/A
2C4RC1K76HR534703	0	3702.5	N/A
2C4RC1K71HR534687	0	3461.2	N/A
2C4RC1K76HR534670	0	3356.2	N/A
2C4RC1K72HR534701	0	3275.7	N/A
2C4RC1K75HR695690	0	3189.1	N/A
2C4RC1K77HR534645	0	3131.5	N/A
2C4RC1K73HR534674	0	3076	N/A
2C4RC1K77HR534662	0	2945	N/A
2C4RC1K71HR534673	0	2887.6	N/A
2C4RC1K78HR695702	0	2833.8	N/A
2C4RC1K76HR797354	0	2778.7	N/A
2C4RC1K77HR534659	0	2679	N/A
2C4RC1K70HR534664	0	2658.6	N/A
2C4RC1K7XHR695698	0	2653.6	N/A
2C4RC1K76HR833785	0	2013.2	N/A
2C4RC1K72HR805465	0	1719.9	N/A
2C4RC1K77HR534693	0	1677.6	N/A
2C4RC1K76HR805453	0	1634.2	N/A
2C4RC1K79HR534677	0	1589.5	N/A
2C4RC1K75HR805461	0	1573.1	N/A
2C4RC1K74HR534652	0	1531.1	N/A
2C4RC1K75HR827265	0	1426.8	N/A
2C4RC1K75HR827234	0	1420.3	N/A
2C4RC1K73HR833808	0	1396.2	N/A
2C4RC1K7XHR833854	0	1385.1	N/A
2C4RC1K78HR695697	0	1344.6	N/A
2C4RC1K70HR534650	0	1299.1	N/A
2C4RC1K70HR534695	0	1266.6	N/A
2C4RC1K73HR833873	0	1168.6	N/A
2C4RC1K7XHR534672	0	1145.4	N/A
2C4RC1K77HR833858	0	1111.8	N/A
2C4RC1K74HR797353	0	1072.6	N/A
2C4RC1K70HR797317	0	978	N/A
2C4RC1K73HR534710	0	971.6	N/A
2C4RC1K7XHR534641	0	945.4	N/A
2C4RC1K72HR824243	0	844.7	N/A
2C4RC1K74HR797322	0	841.2	N/A
2C4RC1K74HR534666	0	839.1	N/A
2C4RC1K73HR797375	0	825	N/A
2C4RC1K72HR833816	0	779.6	N/A
2C4RC1K71HR833841	0	767.7	N/A
2C4RC1K79HR797378	0	727.2	N/A
2C4RC1K70HR797415	0	650.2	N/A
2C4RC1K73HR827264	0	619	N/A
2C4RC1K75HR822194	0	607.9	N/A
2C4RC1K73HR822193	0	547.2	N/A
2C4RC1K73HR833856	0	533.5	N/A
2C4RC1K76HR797399	0	435.5	N/A
2C4RC1K78HR797386	0	429.5	N/A
2C4RC1K78HR534637	0	409	N/A
2C4RC1K71HR833810	0	404.3	N/A
2C4RC1K78HR534699	0	388.1	N/A
2C4RC1K75HR797295	0	370.1	N/A
2C4RC1K74HR827242	0	358.6	N/A
2C4RC1K75HR833759	0	284.5	N/A
2C4RC1K77HR786122	0	279.9	N/A
SADHW2S12M1X06474	0	228.5	N/A
2C4RC1K75HR833857	0	224.4	N/A
2C4RC1K76HR824259	0	213.6	N/A
2C4RC1K76HR805467	0	204.7	N/A
2C4RC1K77HR534676	0	202.1	N/A
2C4RC1K7XHR833756	0	193.1	N/A
2C4RC1K78HR805485	0	185.5	N/A
2C4RC1K71HR822208	0	149.2	N/A
2C4RC1K75HR534658	0	148.8	N/A
2C4RC1K74HR797367	0	144.5	N/A
2C4RC1K79HR786123	0	137.6	N/A
2C4RC1K75HR805508	0	122	N/A
SADHW2S15M1X06484	0	121.5	N/A
2C4RC1K75HR805489	0	114	N/A
2C4RC1K73HR805488	0	108.3	N/A
2C4RC1K78HR833870	0	105.3	N/A
2C4RC1K71HR805523	0	95.7	N/A
2C4RC1K7XHR822207	0	77.9	N/A
2C4RC1K75HR786121	0	71.8	N/A
2C4RC1K71HR822192	0	41.3	N/A
2C4RC1K71HR833838	0	38	N/A
SADHW2S15M1X06468	0	26.4	N/A
SADHW2S17M1602661	0	19.9	N/A
2C4RC1K73HR786120	0	14	N/A
2C4RC1K70HR805481	0	6.8	N/A
2C4RC1K79HR833778	0	3.2	N/A
2C4RC1K77HR833889	0	3.1	N/A
2C4RC1K7XHR833885	0	3.1	N/A
2C4RC1K70HR797365	0	0.1	N/A

Where is the Waymo data on disengagements from their Arizona operations?

Here you go:

https://storage.googleapis.com/sdc-prod/v1/safety-report/Waymo-Public-Road-Safety-Performance-Data.pdf

The data gives accidents both real and avoided by disengagements. The total accident rate was about 1 per 130,000 miles.

 

[Bladerskb]

And in that ~1000 miles he had what looks to be at least 9 disengagements. So using his sample we get 1 disengagement per 111 miles.

It seems like the driving they have done in California isn't really representative of the driving when they are running an actual Robotaxi service...

Where is the Waymo data on disengagements from their Arizona operations?

There's quite a clear difference between disengagement from the car attempting to slam into a wall, run over a ped or crash headon into another car in the opposing lane vs it stopping in a parking lot because its failing to route through a parking lot. I have said this time and time again.

 

[Dutchie]

This is a good video of the current 9.1 FSD beta. The author does not waste your time, every bit of it is informative. He shows everything, warts and all. Personally, I would take this version as shown in a heartbeat. Good enough for me. Not level 5, but close enough that the car could get me through my drives as long as I paid attention.

That is an amazing video. Blown away by FSD’s abilities, especially how it tries to corrects itself passing by the parked truck. I think we get to real FSD sooner rather than later after watching this.

 

[Matias]

But, could additional sensors help the car do better? For example, could a bumper camera help the car see around corners and cross traffic sooner where it does not need to creep forward as much and can make the maneuvers with even higher reliability? Probably. If more sensors could help achieve higher safety, where you can remove driver supervision and trust the car to drive with high confidence, that would be a good thing IMO.

I really hope that Tesla would put cameras to front and rear lights area. It would greatly help and additional cost would be minimal. Hopefully they would also have washers.

 

[jeremymc7]

I really hope that Tesla would put cameras to front and rear lights area. It would greatly help and additional cost would be minimal. Hopefully they would also have washers.

You and me both.

Retrofitting the cameras on the car instead would be easy enough. The additional wiring and load on the hardware is another story.

Maybe in a later car.

 

[Matias]

You and me both.

Retrofitting the cameras on the car instead would be easy enough. The additional wiring and load on the hardware is another story.

Maybe in a later car.

I was talking about future versions. Retrofitting cameras to the current cars would be prohibitive expensive.