The 24 hours part seems open to abuse. Anyone/thing can make an unprotected left at 3-4 am.
1,400 turns with 180 cars is less than 8 turns each in 24 hours. One turn every 3 hours...
(Likely not the whole fleet involved).
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Autonomous Car Progress
- Thread starter stopcrazypp
- Start date
-
- Tags
- Autonomous Vehicles
The 24 hours part seems open to abuse. Anyone/thing can make an unprotected left at 3-4 am.
1,400 turns with 180 cars is less than 8 turns each in 24 hours. One turn every 3 hours...
(Likely not the whole fleet involved).
It you want to be that way:
Is he saying no two turns are alike, but 1,400 are guaranteed to have duplicates?
How many left turns do Teslas perform in 24 hours?
So years ago I remember somebody (maybe Musk?) talking about autonomous cars having a live satellite feed, theoretically it could monitor deer grazing near a highway or maybe supplement autonomous systems in bad weather or poorly marked roads etc. Autopilot with just the cameras sounds like it's going to be pretty good, but I wonder if there are still plans/advantages to adding the satellite feed once they have their network setup? SpaceX just launched a Falcon 9 loaded with Starlink internet satellites
EVNow
Well-Known Member
If every car does one per day, about 400,000 a day
Feels like it’s been very silent the last week. For example check timestamps at:
Andrej Karpathy (@karpathy) on Twitter
My guess is that they want to put some deferred FSD revenue into Q2 and for this they need to complete a delivery milestone, maybe deploying the new network to EAP. And for this they need all hands on deck right now.
Andrej Karpathy (@karpathy) on Twitter
My guess is that they want to put some deferred FSD revenue into Q2 and for this they need to complete a delivery milestone, maybe deploying the new network to EAP. And for this they need all hands on deck right now.
diplomat33
Average guy who loves autonomous vehicles
diplomat33
Average guy who loves autonomous vehicles
Aurora is partnering with Fiat and Chrysler: Aurora and Fiat Chrysler launch self-driving commercial vehicle partnership - Electrek
EVNow
Well-Known Member
That's weird - since Nissan is using MobilEye for their propilot.
Why would that make it weird?
First of all most major car makers have several separate projects on autonomy, especially multi-brand ones (and aren’t they all multi-brand these days).
Second Nissan is using MobilEye for its current consumer car fleet. With Waymo they clearly aim towards the commercial end of the market first and/or are looking beyond the current consumer cars.
Makes total sense to me. MobilEye has a very compelling offering for current consumer cars but Waymo is the current pack leader on commercial autonomy (not so much for current consumer cars) so both have something to offer.
CarlK
Active Member
The difference there though, is MobilEye kicked Tesla out after the Brown incident.
I do not foresee any falling out between MobilEye and Nissan here. The Renault Waymo project is a separate one aiming at different things.
Same with Jaguar, mind you. They are working with Waymo and using MobilEye at the same time — different projects, different products, different scopes and needs.
Apple Confirms Acquisition of Self-Driving Vehicle Startup Drive.ai
I'm dismayed to find that via Testing of Autonomous Vehicles and pages like Autonomous Vehicle Disengagement Reports 2018, that the CA DMV did this:
FWIW, the summary of Drive.ai's results are at UPDATE: Disengagement Reports 2018 – Final Results.
I'm dismayed to find that via Testing of Autonomous Vehicles and pages like Autonomous Vehicle Disengagement Reports 2018, that the CA DMV did this:
Seriously? Before, they had them up, including previous years for years. Now this? I wonder if they're going to back down on this.
FWIW, the summary of Drive.ai's results are at UPDATE: Disengagement Reports 2018 – Final Results.
CarlK
Active Member
Just to note disengagement can be easily gamed to make it look good to impress investors or the public. One only needs to drive on the same familiar route with a few cars over and over again to acquire some impressive data. Those numbers do not mean much until you dig a little deep into how they were administrated.
Last edited:
diplomat33
Average guy who loves autonomous vehicles
Yeah, number of disengagements is somewhat useful in a very broad sense. Obviously, 10,000 miles per disengagements is better than 10 miles per disengagement. But we need more context to fully analyze how good the number of disengagements really is. It's the same problem as Tesla just giving us one number of miles per accident on Autopilot with no context. Without context, it is hard to qualify how good the stat is. In terms of FSD, there are a lot of questions that arise when discussing disengagements. What roads were driven on? Were the miles repeat of the same route or different routes? Were they city streets or highways? Were there a lot of intersections or a lot of main boulevards? What types of disengagements? Disengagements because the car almost hit a pedestrian or disengagements because the car missed an exit or took too long to accelerate on a green light? All those questions will put a different interpretation on the number of disengagements.
CarlK
Active Member
Tesla actually got the most statistically significant data. They came from all owners driving in every part of the country whenever they decide to turn on the autopilot. The only thing is the data need to be compared to non-autopilot data on same roads which I'm not sure if it's done or can be done easily. On the contrary you don't even know if those disengagement data came from just few cars driving on the same stretch of road thousands of times. These are pretty meaningless until details of how they were acquired is known. I'm sure some of those companies or their engineers know how best to game that system. Tesla at least doesn't get to decide when or where you can "test" the autopilot in your car.
diplomat33
Average guy who loves autonomous vehicles
I bet Tesla could give us disengagement data for Autopilot, broken down by type (voluntary disengagement, critical disengagement etc) and road (highway, city street). Considering how many miles the Tesla fleet drives, the data should be statistically informative. Although considering that Autopilot is a work in progress, Tesla may not want to release the numbers just yet. All the disengagements on city streets where AP is not really supposed to be used could bias the results. I do hope that once "automatic city driving" and "traffic light recognition" are released, which is all of the remaining big FSD features, that Tesla would release that disengagement data at some point. Although, they may wait until FSD becomes more reliable so that the data will look better.
Yes, they reported 0 miles and 0 disengagements for many every year but one. In the only year in which they reported CA publoc road autonomous miles >0, they had reported 182 disengagements during 550 miles of autonomous driving that only happened in (IIRC) 2 months of the year.
For me the biggest value in public disengagement data is public announcement of autonomously driven miles — and indeed we have lots of it from pretty much everyone else and nearly none from Tesla.
With Tesla we really don’t know if Tesla has driven any autonomous miles beyond the little that has been seen in public, or are they merely ADAS testing all the time that is a different ball game because the requirements are different.
I get the narrative that Tesla’s ADAS is intended to eventually morph into an autonomous car but testing an autonomous prototype would still be informative of the stage of that progress.
With Tesla we really don’t know if Tesla has driven any autonomous miles beyond the little that has been seen in public, or are they merely ADAS testing all the time that is a different ball game because the requirements are different.
I get the narrative that Tesla’s ADAS is intended to eventually morph into an autonomous car but testing an autonomous prototype would still be informative of the stage of that progress.
Similar threads
- Article
