You need LIDAR

Not sure why this guy still gets engagement. You just spent some time correcting almost every single word he typed. No honesty there
 
Notable phantom braking is like a 1 in 10K mile event at least. Meaning genuine hard braking for something that's not needed. Running a red light has to be a 1 in 100K mile even at least. In a geofenced area I'd think it happens less than that.

That puts the odds of those two things happening at like 1 in 40 million
Out of that 100k usually expressway miles where people usually use ADAS, how many lights did it go through? 2?
 
You are literally just making things up now. Waymo does not exclusively validate with synthetic data. They test in the real world literally every day.

I understand you find it fun and useful to regurgitate press release and twitter content, but outright making things up is new, even for you.
With a small number of vehicles

Even when tesla built the cybercab they had 4000 vehicles for testing

That's not the same thing as running the software on millions of cars

An example of this.

If you want examples of a car overturned at night on the freeway, with no lights on, no cars around, no police, and possibly people walking around that you can kill. You need billions of miles of driving in the middle of the night to get sufficient examples of this.

paying a driver to test during these hours is wasteful because you can't test a lot when there's not much interesting data to collect

Only way to collect sufficient examples of this and validate your system works is to test on millions of consumers
 
Last edited:
Oh fun, new goal posts.

We’re done.
That was never a new goal post. Waymo tests billions of miles in simulation only because they can't get billions of miles through a real fleet.

The point was, at the scale of vehicles waymo has, it's not fast to fix issues.
 
None of these systems are perfect (Waymo, Tesla, Lucid, etc.) There are just too many variables, unpredictable other drivers, weather, road hazards, animals and pedestrians, etc. The key is if they are better than an above average, safe human driver. But if that were the case, I would not have had the 5-7 “accidents” in my 50 years of driving. The data is showing that the Tesla and Waymo systems are better than the average human. We all think that we are better and safer drivers than any autonomous driving system out right now. The key is, if you had a choice, would you rather take an Uber with an unknown human driver or take a Waymo/Tesla Cyber Cab. Based on my limited experience with a few Waymo rides (no Cyber Taxies yet), I will opt for the Waymo.

The key concept is that the average human driver will not improve (and maybe even digress with all the phone and other distractions, etc.) but the autonomous systems will continue to improve. No, they will not be perfect (at least no in our lives) but they will gradually get much better than human drivers.
 
I think the problem here is people think that getting to tesla levels of self driving is easy. It's not. You can't show an improvement of safety with lidar unless you get to the level tesla is today. Which from what we have seen from everyone else not 360 lidar or HD maps is impossible. I've not even seen reaction times from waymo/zoox that appear like tesla, it's just that they see over objects with higher placement of lidar. That of course is a limitation when you run your sensors at 10hz

Waymo for years always made excuses that things that tesla was doing were "not possible" like driving without HD maps, the use of camera only, or the weak compute onboard. All of these things not only happened, but it shows that you can't project your own limitations on what others are able to do. This is not moving a goalpost.
 
None of these systems are perfect (Waymo, Tesla, Lucid, etc.) There are just too many variables, unpredictable other drivers, weather, road hazards, animals and pedestrians, etc. The key is if they are better than an above average, safe human driver. But if that were the case, I would not have had the 5-7 “accidents” in my 50 years of driving. The data is showing that the Tesla and Waymo systems are better than the average human. We all think that we are better and safer drivers than any autonomous driving system out right now. The key is, if you had a choice, would you rather take an Uber with an unknown human driver or take a Waymo/Tesla Cyber Cab. Based on my limited experience with a few Waymo rides (no Cyber Taxies yet), I will opt for the Waymo.

The key concept is that the average human driver will not improve (and maybe even digress with all the phone and other distractions, etc.) but the autonomous systems will continue to improve. No, they will not be perfect (at least no in our lives) but they will gradually get much better than human drivers.
Well got a ride in a new MY uber in France, wow all the distractions on that damn screen, Halloween, Xmas, blah blah, blah. Now I see why they need fsd, so you can see the pretty screen savers
 
Rather than treating HD maps as a strict, brittle track the car is forced to follow, Waymo uses them as a powerful "prior"; analogous to a mental memory of the environment. The onboard computer combines data from its sensors (lidar, radar, and cameras) with the HD map to streamline navigation, especially in low-visibility or complex urban areas.

This setup allows the vehicle's real-time processing power to focus on dynamic or unexpected elements (such as temporary construction, detours, or sudden obstacles) rather than having to re-calculate basic road geometry from scratch every single second.

“Thin chains” are not baked into maps.
The value of having maps in an autonomous driving system ought to be obvious to anyone who has ever tried to drive in an unfamiliar city. It's much easier to drive down a street that you're familiar with, and it takes more focus and attention to drive down a street that you aren't. Particularly if you're in a location that has somewhat different traffic rules than where you normally drive. Generally speaking, those traffic rules are not obvious from just reading signs and road markings as you drive along. Waymo has given a concrete example of where maps help them: figuring out the complicated parking rules in San Francisco.
 
Rather than treating HD maps as a strict, brittle track the car is forced to follow, Waymo uses them as a powerful "prior"; analogous to a mental memory of the environment. The onboard computer combines data from its sensors (lidar, radar, and cameras) with the HD map to streamline navigation, especially in low-visibility or complex urban areas.

This setup allows the vehicle's real-time processing power to focus on dynamic or unexpected elements (such as temporary construction, detours, or sudden obstacles) rather than having to re-calculate basic road geometry from scratch every single second.

“Thin chains” are not baked into maps.

This is false. Waymo premaps speed bumps so you think they don't premap thin chains? I have seen zero evidence that waymo can identify a speed bump in real time and slow down for it. I've seen them go over speed bumps too many times at full speed that are glaringly obvious.

Do you honestly think it takes them 1+ years to launch in a new city, that they are really having to teach the car new driving abilities at this point? No. It's simply refining the detail on the HD maps to fix driving issues.

For 11 years waymo hid their issue of driving in flooded waters, yet in every city they operate there is flooding issues? It's because they premapped these flooding areas in real time.

Waymo literally ran into a telephone pole because the object was assigned a low damage score. That means all objects (even a telephone pole for whatever reason) are being mapped and assigned a damage score

The difference between HD map and not is that with HD map you sync the 360 lidar output. With GPS (and no HD maps) you can be more than 20m off from your actual location.

Waymo is under an NHTSA investigation for failing to stop for school buses. They claimed that "you can't expect perfection." This was after they had been doing unsupervised driving for 10 years. How bad did they perform in 2020 if this is how they perform today? Notice the main issues that waymo has is always related to real-time identification, and never something that can be easily baked into an HD map.
 
Last edited:
The value of having maps in an autonomous driving system ought to be obvious to anyone who has ever tried to drive in an unfamiliar city. It's much easier to drive down a street that you're familiar with, and it takes more focus and attention to drive down a street that you aren't. Particularly if you're in a location that has somewhat different traffic rules than where you normally drive. Generally speaking, those traffic rules are not obvious from just reading signs and road markings as you drive along. Waymo has given a concrete example of where maps help them: figuring out the complicated parking rules in San Francisco.

Everybody uses maps. Tesla uses maps. What you're not getting is waymo/zoox/baidu/etc. use HD MAPS. You take a 360 lidar output, sync the lidar output to the lidar scanned map and you have a centimeter accurate position of where you are within that map. This actually shortcuts the process of doing unsupervised driving since you just drive around with an HD map and dynamically avoid moving objects.

These HD maps are extremely large (Terabytes per city) and they contain a lot more than just lane markings and road signs. They contain details about every object in the environment. Waymo assigns every object a damage score. But there's other details like how fast you should take a curve, potholes, speed bumps, parking lots, etc.

If you are driving without HD maps your GPS can be more than 20 meters off from your physical location, which puts you on the wrong side of the road, past an intersection you are supposed to turn at, etc.

One feature tesla uses is they drive from satellite images, and they plan routes out of a parking lot without relying on maps at all. They use parking lot satellite images to plan the route
 
This is false. Waymo premaps speed bumps so you think they don't premap thin chains? I have seen zero evidence that waymo can identify a speed bump in real time and slow down for it. I've seen them go over speed bumps too many times at full speed that are glaringly obvious.
Please provide a reference why this is false when your own posts tell us it is true. You have told us that Waymo uses 360 degree LiDAR for HD maps which is likely true. I have no inside knowledge of how Waymo works and I suspect you do not either so we do not know for sure. Based on your assumption that the HD maps come from LiDAR, it is fairly obvious that thin chains are not being mapped since LiDAR does not have the resolution to see thin chains. Consistency in you argument would make a better case.
 
Jeremy Clarkson is doing theatre. I enjoy watching theatre but also know that its not always reality. I don't necessarily agree with Clarkson's tactics, as I saw him besmirch the 2008 Tesla Roadster by simulating it run out of charge but saying that quiet part under his breath!
https://topgear.fandom.com/wiki/Tesla_Roadster_Review_Controversy

I did a review of all 3 available autonomous vehicles available in Austin last month:
  1. Tesla CyberCab
  2. Waymo
  3. Tesla Model Y
 
Please provide a reference why this is false when your own posts tell us it is true. You have told us that Waymo uses 360 degree LiDAR for HD maps which is likely true. I have no inside knowledge of how Waymo works and I suspect you do not either so we do not know for sure. Based on your assumption that the HD maps come from LiDAR, it is fairly obvious that thin chains are not being mapped since LiDAR does not have the resolution to see thin chains. Consistency in you argument would make a better case

There are countless videos and anecdotes of waymo running over speed bumps at full speed as if they are not there. I posted one earlier. These are clear as day ones, not like you could remotely confuse them as not speed bumps.

And then what happens is waymo employees map the speed bump (if the car detects an accident happened) and it is subsequently fixed the next day.

Mapping a thin chain simply involves putting something in the map that tells the car that there is a thin chain there. The problem with not using HD maps (as I mentioned earlier) is that your GPS can be 20m off. It's very hard to restrict the car from doing something, and you can't easily just map a pothole and the car goes around it. You would need to both map the pothole but avoid it through real-time detection.

The whole argument against not using HD maps is that it was never possible to drive without them. Tesla (lidar or not) is the only company remotely close to doing unsupervised without HD maps.

Thin chains are seen by lidar, but they are not seen at a confidence level that you can easily slow down for them. There's a lot of noise in a lidar signal, especially in rain, and a thin chain approaches the limit of lidar resolution.

Keep in mind consumer solid state lidars are not 360, lower resolution, and have low FOV. This is not the tech that robotaxis are using
 
Please provide a reference why this is false when your own posts tell us it is true. You have told us that Waymo uses 360 degree LiDAR for HD maps which is likely true. I have no inside knowledge of how Waymo works and I suspect you do not either so we do not know for sure. Based on your assumption that the HD maps come from LiDAR, it is fairly obvious that thin chains are not being mapped since LiDAR does not have the resolution to see thin chains. Consistency in you argument would make a better case.
For anyone interested in a little more information, Waymo has published various blog-style posts about their technology, including a discussion about their mapping. It's aimed at the general audience more than at engineers, so the technical detail is limited. If memory serves, they do map the world around the vehicle using the LIDAR on their cars. Probably using the full sensor fusion model of the surrounding world, although they weren't specific about that. Every deployed car is constantly comparing its sensor readings with the map, and as the world is constantly changing, they do not expect the map to be reliable. When they observe a difference, the car uploads an update based of its sensors, and that is distributed to the whole fleet, and so the map is constantly maintained without them needed to run a separate mapping effort to accomplish that. As far as I can recall, their post was not precise about exactly what's included in the map, and so they weren't specific about things like chains, metal sign poles, fire hydrants, etc. They were clear that the Tesla-fan critique of a Waymo running on pre-determined "rails" is not the case. As anyone who's ever taken a Waymo ride in a city probably could already tell just from watching how it navigates all the obstacles encountered.
 
They were clear that the Tesla-fan critique of a Waymo running on pre-determined "rails" is not the case. As anyone who's ever taken a Waymo ride in a city probably could already tell just from watching how it navigates all the obstacles encountered
Nobody was saying they do this.

it's just that waymo maps fixed objects (i.e. a speed bump) because their argument is probably that these things don't appear overnight

The detail that waymo maps in order to drive is much more than people would like and it's why it takes them 1 year to launch a new city. Mapping is very fast, but refining the maps and fixing issues is not.

The only thing waymo seems to be not good at is real time identification of hazards. For what ever reason they have HD maps with elevation but can't identify flooding streets.
 
Back
Top