I rode in one recently. It was pretty smooth. An interesting thing: the car slowed to a stop in the middle of the street, and I couldn't immediately tell why.

There was a row of parked cars on the roadside, and a person was walking around one of them to open her trunk. I guess the car was able to spot her in the dark and opted to pause. So we were stopped there. The woman threw her hands up as if to say, "whatcha lookin' at?" I had the window down, and I offered the explanation "self-driving car." She said "Ugh, I hate those things." Then the car continued on.

This encounter is very on brand for San Francisco.

Now let's imagine the same scene but with the car stopping right next to thugs breaking into another car :-D

Now that you mention it, robbing a self driving car would be easy. Just stand in front of it and it'll stop for you.

Except if the car decide to drive you (the thug) to the nearest police station, bypassing human input.

I was imagining a self driving car drive by shooting but the car stopped slowed down in an inopportune moment, awkwardly

Inputs like what the humans inside the car are saying should be part of the decision making process. Like everyone inside shooting "DRIVE! DRIVE! DON'T STOP THE Fxxx CAR!". That should lower some thresholds for deciding to stop the car, as with a real human driver.

Being a software developer myself, I would absolutely not want to be a guinea pig (beta tester) inside a 2-ton vehicle that is controlled by software. I know too well how sausage is made.

As someone who used to do AI research, I feel the same way.

It also shows how none of our governments seem to care to prevent companies from testing dangerous products on the street with other unwilling drivers

That's not true with most of the prominent examples you see. All autonomous vehicle fleets in California operate under the approval of the state DMV, and do things like publicize crash reports: https://www.dmv.ca.gov/portal/vehicle-industry-services/auto...

The governments never cared much about how insanely dangerous cars are in the first place and these don't seem a whole lot worse.

Why, what does it improve?

Seriously, it removes a form of social interaction, a type of job and increases the number of vehicles on the road in certain areas again. I fail to see an improvement here. All the money pouring from this is into giant mega-corps as opposed to someone working a 9-5 and driving safely because of the perceived bad behaviour of a few. Defending this AI nonsense as "just a minor problem" is the same as condemming all taxis for "bad drivers" it's just bad logic.

Absolutely! It is so frickin' cool this is happening in the real world right now and it's only going to get better with time!

Fsd isn’t as hard when you have tons of LiDAR cameras, radar, and super high definition mapping.

Tesla is the only one working on real world FSD that doesn’t rely on all that stuff

It’s almost like a problem is solvable when you have the right tools and technology.

It’s easy to solve problems in the lab. Getting it out in the real world with mass manufactured products is the hard part. You cannot scale FSD with high precision mapping and expensive cameras worldwide. Every time there’s a road update or repair, those maps need to be updated.

If you seriously think Tesla is ignorant here, you are not paying any attention. Elon musk’s SpaceX has patents on LiDAR they’ve developed for their rockets, it’s not like they don’t use them and are not aware of their strengths and weaknesses

Robotaxis are out in the real world now. They’re not just in the labs. Your information about mapping is highly outdated. The cars are all self mapping and have the ability to update maps in real time when they detect changes. I think you are the one not paying attention.

Yeah keep dreaming. I’d like to see you put a Waymo car or any other “self driving” car in a city that hasn’t been mapped and see how it does.

And The cars do not update maps in real time, they send any changes they find and the company updates the database after reviewing. If they were able to update in real time then why need a HD map in the first place?

They 100% map and distribute changes to the whole fleet in real time. Waymo and others have talked about it multiple times. It seems you just don’t want to believe it because that would mean your biggest talking point going out of the window.

Send me these talks. I genuinely want to see it, I’m willing to change my opinion on HD maps if it’s true.

Waymo blog talking about change detection and automated information sharing — https://blog.waymo.com/2020/09/the-waymo-driver-handbook-map...

Waymo patent portfolio related to mapping — https://www.youtube.com/watch?v=3uf-5cimWzA

Cruise blog about detecting and rolling out map changes "in minutes" — https://medium.com/cruise/hd-maps-self-driving-cars-b6444720...

As I suspected, I was correct. In the first article you link to, it specially states that when a car encounters something it doesn’t recognize, it reroutes. Why would a car have to reroute if it can just update the map?

What you’re referring to is that the car will send a notice and update the fleet to avoid that area.

Actual changes to the maps including crosswalk changes and other permanent structures require review and update. It is not done automatically.

————— From the article:

If a change in the roadway is detected, our vehicle can identify it, reroute itself, and automatically share this information with our operations center and the rest of the fleet in real time.

We can also identify more permanent changes to the driving environment, such as a new crosswalk, an extra vehicle lane squeezed into a wide road, or a new travel restriction, and quickly and efficiently update our maps so that our fleet has the most accurate information about the world around it at all times.

Why did you not quote the next paragraph from the post?

> *We’ve automated most of that process to ensure it’s efficient and scalable*. Every time our cars detect changes on the road, they automatically upload the data, which gets shared with the rest of the fleet after, in some cases, being additionally checked by our mapping team.

They are literally saying they've automated most of that process and that was 2 years ago. That's completely opposite to what you are claiming. If you don't want to believe they can do it, suit yourself.

Why does the car have to reroute?

To minimize risk? Why would they go into perhaps an area with heavy construction/blockage when they can just reroute? Humans reroute all the time in those instances.

And you’re the one claiming I’m the one grasping at straws lol.

Go ahead and get behind the companies that need training wheels and rail guards. I’m going with Tesla which will be the first true FSD solution

"True FSD" with humans behind the wheel. That's so much better!