I'm the author of this. It was certainly surprising to learn that a lot of what AVs are doing right now is limited by the amount of power available to the computers. It didn't make it into the article in full, but we were told that in general more computing power means detecting and classifying more object types, tracking those objects over longer periods of time, and doing more detailed motion prediction and planning across longer time horizons. They were explicit that the power they can squeeze out of the vehicle's alternator is currently a significant limiting factor to the performance of the system.

I am not a car specialist, but I am not buying this. Alternator can be upgraded: https://www.trucknews.com/products/600-amp-alternator-for-mi...

With 600 amps@24V one has 10 kW power for computers and the truck. That’s what 300 decent graphic cards consume.

I immediately thought about the car stereo guys they run stronger alternators and power massive amount of subs and speakers. But I know nothing about big trucks and their electrical needs. They have refrigeration and other electrical needs I am unsure of.

You made me curious. Big refrigerators and shipping containers have their own diesel engine: https://en.m.wikipedia.org/wiki/Refrigerator_truck They are independent from big truck’s power system. That is a solution for powering electronics - external diesel generator.

10 kW is almost 10% of the max output of the engine in a honda civic. for another comparison, that would be like running three typical car AC systems at max simultaneously. so not infeasible, but a meaningful hit to fuel economy in a smaller vehicle. it would be even more detrimental to the range of a small EV.

If so, why not use cars without alternators? 48V mild-hybrids have much higher capacity starter-generators instead of alternators, and electric cars can generally pull power from their batteries at the same rate they can charge it. For example, for a Tesla Model 3 that's 250 kW.

Hybrids are absolutely the most effective way to go here in the near term, there is just no question in my mind. I'm honestly very, very surprised we haven't seen that take off yet-- either it's a lot harder than it looks, or owners of large fleets are irrationally gunshy. I've been expecting it to show up and start kicking ass any minute now for at least the last 7 years.

Seems like electric vehicles are not limited by this problem of power availability. One just has to be concerned by how much that power consumption for computer operation affects vehicle range.

People also tend to give trucks a wider berth, or at least notice you more and respond more readily to your actions.

I’ve only ever owned small cars, and in the handful of times I’ve needed to drive a U-haul truck or similar the trip started with trepidation. But I quickly discovered that other drivers have a strong tendency (self-interested, of course) to help big trucks out.

>"Trucks are also a better platform for autonomy, with their large size providing more power for computers"

I'm no expert but "power for computers" doesnt strike me as a bottleneck, Geohot/Comma seem to be doing fine, this just sounds like marketing bs

Trucks tend to have bigger alternators, so there is more power available. Even the most basic economy car has more than enough engine power, but the alternator can't deliver that power. The engine power instead mechanically goes through the transmission to the wheels. (in a hybrid car there are two alternators, one for the wheels and when for the rest of the electronics so the same problem exists)

the is solvable. you can buy bigger alternators, and in some cases there is enough room that they will even fit... However it is an additional upgrade that needs to be considered.

I don't disagree with what you're saying, but my point was that I don't know if power consumption is even relevant, considering what Comma are doing from a phone on a windshield. I don't know (I'm not an expert) but I don't think putting TPUs in trucks is a requirement for solving driving.

That's right the limitation is Data. You can't effectively use the extra compute if you don't have enough data to train a bigger neural network. https://jperla.medium.com/why-tesla-autopilot-ought-to-be-aw...

Do you actually have data to back that up? I read your article, which (no undue disrespect intended) looked to be full of unsubstantiated qualitative data/graphs and opinion. Are you an "expert"? Or are you trying to get medium clicks on your article/ purchases of your book?

I built self driving cars (at Lyft), including behavioral planning algorithms.

I have run these essays by engineers at Waymo, Lyft, Uber, Nuro, etc and they have verified all of the information (I've incorporated any feedback).

That's fair, although credentials != proof. Working for Waymo, Lyft, Uber, Nuro etc is a credential. The engineers you mentioned have verified that what you wrote fits with the method/ approach they think will work. There is a possibility that they're/ you're wrong.

(This isn't intended as a personal attack)

You asked for credentials...

Yeah fair, I wanted to see if you were full of shit, but you probably know more than me, and bickering and points scoring over stupid things is how hacker news turns into reddit, sorry.

This is true, but the size of the vehicle is probably correlated to the scale of any failure modes.