Is Tesla In A Lose-Lose State of affairs With Autonomous Driving?

Help CleanTechnica’s work by way of a Substack subscription or on Stripe.

One in every of our glorious commenters posted an intriguing remark this morning below an article about how robotaxi app utilization is break up throughout the US market. I’ve been following and engaged on this debate for greater than a decade, and I don’t assume I’ve ever seen a remark fairly like this. It’s actually an eyebrow raiser and will get you pondering. Learn the remark beneath, and be happy to ship in your individual full article response if in case you have one — professional or con. I’ve a sense this intriguing piece from “Matthew2312” will encourage extra lengthy takes on the place issues are headed on this market. —ZS

Let me provide an statement: If Tesla is true about autonomy, Tesla loses. If Tesla is incorrect about autonomy, Tesla loses extra. Right here is the very truncated motive why:

If Tesla is true about autonomy…

Tesla is arguing that every one autonomy requires is a middling computational stack like HW4 plus a handful of middling decision cameras, and good software program. Whether it is proper, there are already hundreds of thousands of automobiles on the highway with considerably extra succesful {hardware} than what Tesla has fielded. (My gen 2 R1S with 11 mp cameras and twin Orin processors plus 360 radar is a fantastic and never significantly particular instance.) As soon as Tesla proves it may be executed, the race shall be on to copy the Tesla software program on present {hardware}. It’s just like the Wright Brothers, Kitty Hawk didn’t construct a “moat,” it opened the aggressive floodgates.

How lengthy to copy as soon as that is confirmed? Not lengthy. Greater than a yr. Lower than three. The coaching information is broadly obtainable. The coaching infrastructure is on the market. Tesla can have confirmed the template. The world is aware of learn how to be taught from a confirmed system. Rack 100 FSD computer systems, feed them artificial sensor streams from a simulator, file the outputs, and you’ve got a dense behavioral dataset from the one system that’s demonstrated the aptitude — no supply code required. Mix that with broadly obtainable coaching information and commodity coaching infrastructure, and replication is a bounded engineering downside, not an open analysis query. The monetary incentives shall be excessive as a result of all these firms can get a pair thousand {dollars} from already present automobiles and that may be a once-in-a-generation alternative to seize a couple of billion {dollars} in additional margin.

The implication is clear. If Tesla is true, the business will rush in and retrofit hundreds of thousands of automobiles with Stage 4 autonomy. All of them will be taxis (be aware for file a three-year-old R1S is a WAY cooler robotaxi than a two-door cybercab). Actually each single OEM is able to instantly transport “L4 Prepared” automobiles beginning within the present mannequin yr. It will likely be the quickest commoditization since… effectively since Kitty Hawk.

If Tesla is incorrect…

It has a large legal responsibility overhang. The remainder of the business laps it with the multi-sensor fusion methods. They develop into an also-ran within the robotaxi business (which might be not that large a deal in actuality because the enterprise has mediocre returns, however it’s a killer within the brief time period).

Be aware this: if Tesla is true — everybody already has L4 {hardware} deployed of their present and future fashions. If Tesla is incorrect, just one firm has to begin over and re-engineer its total method to autonomy.