In the piece — titled “Can You Fool a Self Driving Car?” — Rober found that a Tesla car on Autopilot was fooled by a Wile E. Coyote-style wall painted to look like the road ahead of it, with the electric vehicle plowing right through it instead of stopping.
The footage was damning enough, with slow-motion clips showing the car not only crashing through the styrofoam wall but also a mannequin of a child. The Tesla was also fooled by simulated rain and fog.
It’s also very expensive.
Relative to the car, it’s peanuts at large quantities
Sorry but I don’t get it. You can getva robot vacuum with lidar for $150. I understand automotive lidars need to have more reliability, range etc. but I don’t understand how it’s not even an option for $30k car.
You’re car’s not driving indoors at 1mph with the maximum damage being tapping but not marring the wall or vehicle.
You need high speed, bright lasers, and immense computation to handle outdoor, fast, dangerous work
The is Elon we’re talking about. Why pay a few hundred bucks to improve safety when it’s cheaper and easier to fight the lawsuits when people die?
he does not want to pay $1 for rain sensors and $2 for ultrasonic parking sensors, any price for lidar must be unacceptable
They were much more expensive years ago when the decisions were made to not use it. Costs have come down a lot. And cars can have more than 1 if you’re going to use it. That also means more compute needed so a stronger computer and more power draw meaning less milage, which means bigger battery for same mileage. It all adds up.
Edit: might even impact aerodynamics, which again means more battery, which is more expensive.
The power draw to process the LIDAR data is negligible compared to the energy used to move the car. 250-300 Watt hours per mile is what it takes to move an electric sedan on average. You might lose a mile of range over an hour of driving, and that’s if you add the LIDAR system without reducing the optical processing load.
LIDAR sensor housing can be made aerodynamic.
While it’s true that LIDAR was more expensive when they started work on self-driving, it doesn’t make sense for them to continue down this path now. It’s all sunk cost fallacy and pride at this point.
A mile per hour is probably about right, but that’s probably per lidar. Waymo has 4 for example, so on a 300mile vehicle that could be 17 miles at 70mph.
Even if you can make it aerodynamic it’s still not going to be as aerodynamic as it not being there.
Sunk cost fallacy make sense, but I’d say it’s also the fear of the massive lawsuit/upgrade cost if wrong due to his statements.
I tried to look up how much power these self driving systems are pulling, but it looks like that will require a deeper dive. The only results I got from a quick search were from 2017-2018, and the systems were pulling around 2 kW. I’m sure that’s come down in the 7-8 years since, but I don’t know how much.
I think you’re right on the lawsuit/upgrade cost. They are on the hook to supply Full Self Driving to all the buyers who bought the option. It’s clear they’re not going to be able to provide it. It looks like there are several class-action lawsuits currently underway.
I think the older Tesla system (HW3) was around 300w, but I think the newer system is more now as they beefed up the compute, but I haven’t seen a number on that. The old system is pretty much maxed out though with no room to grow other then making things more efficient vs just more raw power usage.
A lot of the older hardware back then wasn’t purpose built for driving and was more repurposed general graphical compute, so it was less efficient hence the 2Kw you were seeing. Tesla built ASICs for the driving computer to bring costs and power usage down.
With the newer purpose built Nvidia stuff I’m sure that has brought the power draw down a lot though, likely relatively close (better or worse I don’t know) than Tesla’s watt per performance.
edit: clarity
IIRC robot vacuums usually use a single Time of Flight (ToF) sensor that rotates, giving the robot a 2d scan of it’s surroundings. This is sufficient for a vacuum which only needs to operate on a flat surface, but self driving vehicles need a better understanding of their surroundings than just a thin slice.
That’s why cars might use over 30 distinct ToF sensors, each at a different vertical angle, that are then all placed in the rotating module, giving the system a full 3d scan of it’s surroundings. I would assume those modules are much more expensive, though still insignificant compared to the cost of a car sold on the idea of self driving.