[natch]

>There are at least 3 known cases where autopilot failed and someone died.

Depending how you falsely define "failure," Autopilot can be falsely said to fail all the time.

If you set an unrealistic bar, it can fail each and every second of every drive, according to some fallacious definitions.

Fully replacing the driver is not a claimed feature of the current system.

Future systems, sure. But current systems? Nope.

Therefore, as has been clearly stated in many places, the human driver is responsible for maintaining safe operation of the vehicle.

So, the human is to blame in these cases, not Autopilot.

[deanak]

Why are you getting into the definition of fallacious? Read the NTSB report.

> The Tesla Autopilot system did not provide an effective means of monitoring the driver’s level of engagement with the driving task, and the timing of alerts and warnings was insufficient to elicit the driver’s response to prevent the crash or mitigate its severity. Requirements are needed for driver monitoring systems for advanced driver assistance systems that provide partial driving automation (SAE Level 2 systems), and Tesla needs to develop applications that more effectively sense the driver’s level of engagement and that alert drivers who are not engaged...

> Despite the system’s known limitations, Tesla does not restrict where Autopilot can be used. Tesla should incorporate system safeguards that limit the use of partial driving automation systems (Autopilot) to those conditions for which they were designed...

> The Tesla’s collision avoidance assist systems were not designed to, and did not, detect the crash attenuator. Because this object was not detected, (a) Autopilot accelerated the SUV to a higher speed, which the driver had previously set by using adaptive cruise control, (b) the forward collision warning did not provide an alert, and (c) the automatic emergency braking did not activate. For partial driving automation systems to be safely deployed in a high-speed operating environment, collision avoidance systems must be able to effectively detect potential hazards and warn of potential hazards to drivers

Autopilot 1) failed to detect a distracted driver and engage them, 2) is not limited by Tesla to areas where it actually works, and 3) depends on collision avoidance which does not work. Tesla is claiming they will solve these problems without LiDAR. Volvo, Waymo, GM, VW, and many more claim that even semi-autonomous driving isn't possible without it. We'll see how many more people have to die before Tesla changes their mind.

[natch]

We have to be careful about moving the goalposts to measure Autopilot according to supposed qualities it never claims to have.

It’s likely that implementing the suggested solutions would add more problems than it solves. For one thing some of them would exacerbate one problem you seem to be concerned about, driver overconfidence and over reliance on the system.

So while these bureaucratic opinions are fascinating, like many such committee driven conclusions they should be taken with a grain of salt. Tesla has thought things through pretty well.

Ultimately the best solution for now imho is to keep full responsibility with the driver.

[deanak]

If autopilot allows itself to be activated in an area where it doesn't work, who is accountable for that? If it depends on collision avoidance that performs far worse than other solutions, who is accountable for that? When Tesla sells a $10k package that calls itself autopilot and then people die when they have it turned on, who is accountable for that?

> So while these bureaucratic opinions are fascinating, like many such committee driven conclusions they should be taken with a grain of salt. Tesla has thought things through pretty well.

Every other driver assist technology limits its own scope to prevent over-reliance. That's because those companies (Subaru/Toyota/GM/Ford/Volvo/BMW/VW) are working with regulators across the world to develop and adhere to standards. If Tesla's solution was better, you'd have some numbers to back it up instead of a bizarre assertion that industry experts are suddenly clueless when they are part of a regulatory body.

Tesla has created an unreliable semi-autonomous system and made it available before it's ready against the advice of the NTSB and other regulators. That is irresponsible and people who would have otherwise been paying attention have died because they, like you, assumed that "Tesla has thought things through pretty well."

Much like an IT department that fights tooth and nail against a third party security audit, it should be obvious why Tesla is so dedicated to pretending that there isn't a problem. They don't want to admit autopilot isn't ready, and will likely never be dependable until they have proximity sensors in addition to cameras like every other solution does. They'd rather blame the drivers for relying on the technology they just paid for. It's a ridiculous defense that will ultimately fall apart.

[natch]

You don’t seem to be reading the clear onscreen warnings about this stuff from Tesla. There is a lot of misinformation out there, so I can understand how you could form negative opinions about them. My personal, first hand experience of their stance, as the owner of two Teslas both with the FSD option paid for, has been different from what you describe.

BTW about FSD I don’t expect it anytime soon. I paid for it (yes twice!) to support the company at a time when it was in a fragile financial state. It does give me some small benefits for now but I do look forward to more in the future. I don’t know how it will turn out but I’m thrilled to help do a small part to enable awesome futuristic technology! And to support a company that is not brain-locked sticking to the playbook of a committee of fuddy duddy old world car thinkers.

[natch]

>Tesla has created an unreliable semi-autonomous system and made it available before it's ready

Can you think of a better way?

Really. The obvious (bad) answer people come up with here is "just wait until it's perfect and release it then, and only then."

But I don't see how that's realistic. How, in your mind, could that possibly work?

[deanak]

There are plenty of technologies that are working in a limited scope for driver assist, and functioning reliably if not perfectly. Tesla's death rate (measured in vehicle years) is three times that of their competition in the luxury segment.

Step 1 is to not sell something called Full Self-Driving/Autopilot when it can't do either of those things. Step 2 is to develop a reliable system (per NTSB advice) to make sure the driver is paying attention. Step 3 is to make sure it's only active in the domain where it can be trusted. Step 0 is to not do anything else until your collision avoidance works as well as your competitors.

Consider these differences:

"Subaru EyeSight Driver Assist Technology" -- with disclaimer about not being optimal in all conditions

"GM SuperCruise hands-free driving-assistance" -- with a similar disclaimer

"Tesla Full Self-Driving" -- and their disclaimer is "Full Self-Driving is in early limited access Beta and must be used with additional caution. It may do the wrong thing at the worst time, so you must always keep your hands on the wheel and pay extra attention to the road. Do not become complacent."

The marketing bait and switch is pretty common, but this is "Thanks for the $10,000 USD for Full Self Driving. It doesn't work. Don't trust it. In fact, pay extra attention while it's on."

I have never seen the tech community so excited about paying to be alpha testers for technology that is literally killing its users.

[natch]

I understand what you’re saying but you missed the question. It was how do you develop an FSD system? The systems you mention are not FSD, and only one of the three (Tesla) is working toward FSD. Do you see a better approach to get to FSD?

I don’t agree that the system is killing its users of course. That kind of inflammatory wording doesn’t help anything imho. The users are possibly contending for Darwin awards… they are doing it to themselves.