hph6203 said:

---

harge57 said:

---

The lidar system is not a significant cost factor in my opinion and ultimately provides better results. I think at scale the FSD systems in the future will all have a lidar component.

Try defending a FSD accident in court when not using lidar.

---

Assumes LiDAR will durably provide improvement over an advanced vision only system, which is not necessarily true. In the short term Waymo is targeting a vehicle in the ~$80,000 range, Tesla in the <$20,000 range. LiDAR is not the only thing Waymo is relying upon to create their self driving technology that contributes to its increased expense over Tesla's process.

---

It does not have to be an improvement. It is a tool in the tool kit for those cases when CV has issues.

In the hypothetical where a pure vision system makes a fleet wide accident causing error five times every 10,000,00 miles and a LiDAR + Vision system makes 7 errors per 10,000,000 miles it is important that the LiDAR exist in the system, because while it increased the overall error rate, it reduced the vision error rate from 5 to 4 while introducing 3 errors originating from the LiDAR data?

In a very simplistic view yes. It it is the FIT (failure in time) of the entire system including redundant checkers that matters. There are various kinds of checkers, but one of the key types is algorithmic diversity. Algorithmic diversity protects against systemic problems in s specific algorithm. LIDAR or RADAR are simple object detection and ranging that is very different from CV object detection and ranging.

One other key thing that is separate from failure rate is failure detection and response. Just because a failure is rare, it does not mean you do not need to detect an respond to these failures. When doing proper safety analysis you need to list out the failure modes of each system, how they are detected and how the responds in the case of these failure. Cameras can be a multiple point failure because there are many ways they can be obscured so handling of that would be a hard requirement.

Eventually when FSD gets approved, and there is an accident from FSD then you get to sue the car owner/their insurance AND NOT Tsla (gov approved). No different than the current scenerio.

If AI/Compute can run 1 Mil scenarios of similar situations, then it will make better decisions than humans in a fraction of the time with quicker reflexes.

AI/Computer currently wins in almost all contests against a human, so why not driving?

Eventually data will show that FSD will be many standard deviations better than even the best human drivers causing a fraction of the accidents. The accidents will be more minor compared to someone sleeping/drunk/impaired at the wheel.

I think we are 1 years away from some city approving full Robotaxi and data will show that it is much much safer.

Medaggie said:

---

Eventually when FSD gets approved, and there is an accident from FSD then you get to sue the car owner/their insurance AND NOT Tsla (gov approved). No different than the current scenerio.

If AI/Compute can run 1 Mil scenarios of similar situations, then it will make better decisions than humans in a fraction of the time with quicker reflexes.

AI/Computer currently wins in almost all contests against a human, so why not driving?

Eventually data will show that FSD will be many standard deviations better than even the best human drivers causing a fraction of the accidents. The accidents will be more minor compared to someone sleeping/drunk/impaired at the wheel.

I think we are 1 years away from some city approving full Robotaxi and data will show that it is much much safer.

---

So the Telsa auto pilot failure were it drove into the side of a truck because it thought it was sky is minor. This is the decision you have to protect against.

As for suing the owner that would be terrible legal precedent. The owner has no control of the FSD.

Safer yes when everything works correctly. The problem is electronics have failures. These can be due SW defects, HW defects, environmental factors or just random failures most computer systems have. These systems need to be resilient to these types of failures.

Here's what my wife's Model Y saw when we pulled into the garage a while back...
Not quite ready for primetime? We don't use FSD, but... just sayin'

The GUI isn't a representation of what the car actually perceives. It's placing premade assets on the screen that as closely represent what it sees based upon the available assets. "Garage" isn't in the asset repository, so you got a couple of semis.

You seemingly don't have it available on your car, it's only on cars made after ~October 2022 without ultrasonic sensors, but the High Fidelity Park Assist is a better representation of the cars perception. It's awesome and accurate.