Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Autonomous Car Progress

This site may earn commission on affiliate links.
How come?
How come you are so dishonest and don't disclose your conflict of interest ?

Are you getting paid by a Tesla competitor ? If so disclose it - no, you don't have to tell us you work for GM, just disclose that you work for a competitor.

Oh - you think that will make your posts less credible ? Thats the whole point - you don't want to disclose your conflict of interest to unethically gain credibility.
 

Full release notes:

Expanded driving capabilities

  • We expanded the AV’s ability to drive in foggy and wet weather with the addition of moderate fog and rain by shipping perception improvements (PSeg v5.1.4 and PD v28) that improve object footprint accuracy by 45% in fog and increase precision of object detection around puddles and tire spray by 44%. We also enabled our AVs to adjust their speed dynamically as the level of fog increases without needing to pull over as often. In addition we shipped a wet lens detection model that allows each AV to detect when camera lenses are wet and need to be cleaned.
  • We enabled the AV to turn left from mid-block side streets onto one-way major roads. This opens up about 400 new intersections in SF that the AV can now route through.
  • We improved the AV’s contextual scene understanding of active emergency scenes and proactively re-route to avoid these scenes.
Improved core driving behavior

  • Shipped TSEL v12 which improves lane change behavior by changing lanes more proactively to continue along the planned route, being more decisive to change lanes and braking more smoothly (+21%) during lane changes.
  • Shipped AV seeds v4 which increases the quality and diversity of potential AV trajectories considered, resulting in 3%-5% smoother driving, especially while making lane changes.
  • Shipped STA-B v11 which improves behavior around motorcyclists by 26% by introducing fine-grained classifications and training data focused on motorcyclist behaviors in key scenarios.
  • Shipped LRD v2 which increases the precision of nearfield object detections and further increases detection of nearfield pedestrians in non-standing poses.
  • Shipped LPC V6 which increases steering and braking comfort and improves stopping behavior at stop signs and traffic lights.
  • Shipped DLA v2 and KSE v15 which combine to increase detection of high-speed vehicles by 15%, and improve speed estimation of high-speed vehicles by 70%.
  • Increased lateral maneuvering capabilities for expanded obstacle avoidance optionality as well as improved behavior on reduced friction surfaces and higher speed roads.
Increased reliability

  • Significantly expanded core remote advisor capabilities in large intersections, at higher AV speeds, when dealing with stuck situations, closed roads, debris in the roadway or narrow gaps, resulting in fewer trip interruptions.
  • Shipped SURF+ which provides remote advisors with more fine-grained ability to assist with the movement of the AV.
  • Increased redundancy in the controls system, enabling the AV to pull over and stop in a more humanlike way in the rare case of certain diagnostic faults.
  • Decreased stuck scenarios during pickup and dropoff due to doors left ajar by passengers.
Improved rider experience

  • With this release, we enabled the AV to now make sharper turns and cross bike lanes to pull over, reducing walking distance on 8% of trips, especially near the Embarcadero in SF.
  • Increased available pullover locations by 10% in SF, resulting in 3.5% less walking distance for riders during pick up and drop off.
  • Increased the lateral distance between the AV and other parked cars during pick-up and drop-off to provide more space for riders to enter and exit the AV.

 
  • Informative
  • Like
Reactions: GSP and powertoold
How come you are so dishonest and don't disclose your conflict of interest ?

Are you getting paid by a Tesla competitor ? If so disclose it - no, you don't have to tell us you work for GM, just disclose that you work for a competitor.

Oh - you think that will make your posts less credible ? Thats the whole point - you don't want to disclose your conflict of interest to unethically gain credibility.

I don't think conflict of interest matters here, as many of us are long TSLA.

Regardless of conflict, I think we can come to our own conclusions based on facts and reason. Some of us aren't as good in reasoning, and nothing can change that, so it's pointless to change the way people think.
 
  • Disagree
  • Like
Reactions: DanCar and EVNow
After watching some of the videos of Cruise creeping through intersections and behaving in ways that other drivers then give a wide berth, those safety figures don't surprise me.

When people slow down and leave lots of room between them and others, collisions go down.

Obviously, the benefit of all AV will be the vehicles will be able to pack in and move in sync with each other, thus providing the smooth movement through a city that current humans believe they are getting by insisting on leaving no room between them and the car in front lest someone 'cut in' and slow them down by a fraction of a second.

Tesla's requirement of leaving more car lengths is an example of how, if we all drove that way, there'd be fewer collisions with fewer slowdowns for the rest of the traffic as they funnel past. As well, it is better and most times faster to move steadily at 30 mph than going between stopped and 60 and back to stopped. Uses less fuel as well (be it petrol or electricity).
 

Full release notes:

Expanded driving capabilities

  • We expanded the AV’s ability to drive in foggy and wet weather with the addition of moderate fog and rain by shipping perception improvements (PSeg v5.1.4 and PD v28) that improve object footprint accuracy by 45% in fog and increase precision of object detection around puddles and tire spray by 44%. We also enabled our AVs to adjust their speed dynamically as the level of fog increases without needing to pull over as often. In addition we shipped a wet lens detection model that allows each AV to detect when camera lenses are wet and need to be cleaned.
  • We enabled the AV to turn left from mid-block side streets onto one-way major roads. This opens up about 400 new intersections in SF that the AV can now route through.
  • We improved the AV’s contextual scene understanding of active emergency scenes and proactively re-route to avoid these scenes.
Improved core driving behavior

  • Shipped TSEL v12 which improves lane change behavior by changing lanes more proactively to continue along the planned route, being more decisive to change lanes and braking more smoothly (+21%) during lane changes.
  • Shipped AV seeds v4 which increases the quality and diversity of potential AV trajectories considered, resulting in 3%-5% smoother driving, especially while making lane changes.
  • Shipped STA-B v11 which improves behavior around motorcyclists by 26% by introducing fine-grained classifications and training data focused on motorcyclist behaviors in key scenarios.
  • Shipped LRD v2 which increases the precision of nearfield object detections and further increases detection of nearfield pedestrians in non-standing poses.
  • Shipped LPC V6 which increases steering and braking comfort and improves stopping behavior at stop signs and traffic lights.
  • Shipped DLA v2 and KSE v15 which combine to increase detection of high-speed vehicles by 15%, and improve speed estimation of high-speed vehicles by 70%.
  • Increased lateral maneuvering capabilities for expanded obstacle avoidance optionality as well as improved behavior on reduced friction surfaces and higher speed roads.
Increased reliability

  • Significantly expanded core remote advisor capabilities in large intersections, at higher AV speeds, when dealing with stuck situations, closed roads, debris in the roadway or narrow gaps, resulting in fewer trip interruptions.
  • Shipped SURF+ which provides remote advisors with more fine-grained ability to assist with the movement of the AV.
  • Increased redundancy in the controls system, enabling the AV to pull over and stop in a more humanlike way in the rare case of certain diagnostic faults.
  • Decreased stuck scenarios during pickup and dropoff due to doors left ajar by passengers.
Improved rider experience

  • With this release, we enabled the AV to now make sharper turns and cross bike lanes to pull over, reducing walking distance on 8% of trips, especially near the Embarcadero in SF.
  • Increased available pullover locations by 10% in SF, resulting in 3.5% less walking distance for riders during pick up and drop off.
  • Increased the lateral distance between the AV and other parked cars during pick-up and drop-off to provide more space for riders to enter and exit the AV.

reading this made me realize one of the weaknesses of the current Tesla system.
With FSDb, the navigation routes for a human driver without bothering to validate against FSDb abilities. Whereas Cruise limits the routing based on its ability to navigate.
 
  • Like
Reactions: DanCar and JHCCAZ
With FSDb, the navigation routes for a human driver without bothering to validate against FSDb abilities. Whereas Cruise limits the routing based on its ability to navigate.

I am not sure I would call it a weakness per se. It is the different approach between Tesla and Cruise:

FSD beta is intended as a driver assist that the driver can use everywhere. So Tesla does not want to limit routes. Also, there is less of a need to validate a route against FSD beta's abilities since the driver is expected to take over if FSD beta can't handle something. It would only be a weakness if Tesla's goal was to always have zero interventions but that is not the case. Since FSD Beta is a driver assist, Tesla actually does not really care if the driver has to intervene sometimes. Cruise is doing a driverless robotaxi service. There is no driver supervision. If the Cruise AV can't handle something, it can result in a stall or incident. So Cruise has every incentive to limit routes to what the system can handle better. Additionally, a robotaxi service does not necessarily need to route everywhere, it just needs to get the passenger to their destination. So a robotaxi can select the easier route that still gets the passenger to their destination.
 
Rob of TeslaDaily pointed out yesterday that Cruise uses a much different standard rate than Tesla. 20k miles per crash (50 per 1 Million miles) compared to NHTSA which shows 1 collision per 500,000+ miles.

This is what I've been saying for sometime - the definition of collision is different. Can't compare.
 
  • Helpful
Reactions: GSP
I don't think conflict of interest matters here, as many of us are long TSLA.

Regardless of conflict, I think we can come to our own conclusions based on facts and reason. Some of us aren't as good in reasoning, and nothing can change that, so it's pointless to change the way people think.
And it's good to discuss the topic versus trying to attack posters. Frustration tends to drive us to that point. Unfortunately I fall into that trap with the online shills. Oy!
 
  • Like
Reactions: GSP and diplomat33
reading this made me realize one of the weaknesses of the current Tesla system.
With FSDb, the navigation routes for a human driver without bothering to validate against FSDb abilities. Whereas Cruise limits the routing based on its ability to navigate.
The nav system in a Tesla plans a route before the car knows that FSDb will be tasked to accomplish it. So, if the nav system avoided situations that are difficult for FSDb, the route may not be optimal for hand driving.

In theory, Tesla's could have a setting for the nav system that could control this, but, given that FSDb operates over a far, far larger road network than Cruise (or Waymo), generating and maintaining a set of problem areas for the nav system might become rather cumbersome.
 
The nav system in a Tesla plans a route before the car knows that FSDb will be tasked to accomplish it. So, if the nav system avoided situations that are difficult for FSDb, the route may not be optimal for hand driving.

In theory, Tesla's could have a setting for the nav system that could control this, but, given that FSDb operates over a far, far larger road network than Cruise (or Waymo), generating and maintaining a set of problem areas for the nav system might become rather cumbersome.
Plus that’s not their design objective. They don’t want a car that can get you nearly everywhere you want to go, they want a system that can drive anywhere a human can, and having FSD fail and be corrected by human (customer) drivers is the entire way they test and refine the system. They wouldn’t solve the problem by hiding from it.
 
  • Love
Reactions: GSP
I've seen some discussion about Mobileye recently, and I'm just here to say that Mbeye is nowhere near Tesla for achieving 2x human safety fsd.

I try to repeat this every few months, but it's kinda pointless because it requires my level of intuition and judgment to understand it, and if you're not understanding things the way I do, you'd have a different perspective.

Mbeye continues to be a joke in my view. People who know know.
 
  • Disagree
Reactions: diplomat33
IMG_5073.jpeg

(Sorry if I missed this being posted already; it’s a few days old.)
 
It's basically a FUD hit piece.
Please elaborate.

I saw some quotes from people who preferred the autonomous vehicles.

I liked the quote “Move Fast And Hit Brakes On Things.”

To me it seems a relatively reasonable presentation, but low on any detail about any of the actual issues. It sounds to me like the cars sometimes really suck and are frustrating to other drivers. I think on that we can all agree?
 
  • Like
Reactions: stopcrazypp
Please elaborate.

I saw some quotes from people who preferred the autonomous vehicles.

I liked the quote “Move Fast And Hit Brakes On Things.”

To me it seems a relatively reasonable presentation, but low on any detail about any of the actual issues. It sounds to me like the cars sometimes really suck and are frustrating to other drivers. I think on that we can all agree?

Yes, we can agree that Cruise and Waymo vehicles sometimes get stuck. I think I was thinking of a different article that was super critical of a Cruise ride. This Wash Post article is not FUD. Although, I think it does focus a bit too much on the negatives.
 
Last edited:
Please elaborate.

I saw some quotes from people who preferred the autonomous vehicles.

I liked the quote “Move Fast And Hit Brakes On Things.”

To me it seems a relatively reasonable presentation, but low on any detail about any of the actual issues. It sounds to me like the cars sometimes really suck and are frustrating to other drivers. I think on that we can all agree?
Yep, I read it and it was a balanced take. Gave every player a voice basically, from public that were negative about it, public that were positive, the city's take, the companies' take, and the riders' take. There's not a lot of details, but basically laid out most of the positive and negative things about them.