A certain amount of problems are a fact of life. Do you have a particular reason to think the amount will be overwhelming here?
The main thing is that self driving semis are fine if it's going through Nevada or death valley or some other sparsely driven road that doesn't have any intersections. But once the semi enters a suburb or big city...it's going to be an issue.
And Tesla just seems rather unserious. It's one thing to miss a prediction several years out, it's quite another to say you're confident "this year" or "next year" with specifics, repeatedly, and constantly miss by miles. Not to mention the obsession with having fewer sensors. Teslas with the FSD beta still routinely screw up basic driving tasks, not edge cases, like deciding to try and run into a cement wall straight ahead.
Semi drivers do that 220 miles (and then some), each and every day, on quiet interstates and havoc strewn city highways.
And we’re also talking big rigs here with significantly longer stopping distances and far less maneuverability.
And they have to deal with the vagaries of weather. A fully automated rig is no good to anyone if the software decides to not proceed with no driver to override it.
The shift in attitude towards Tesla here and elsewhere is really funny to me because it coincides more or less with the time period in which Tesla started to actually deliver the things they had promised years ago.
The fact that after more than a decade of development, Teslas still have trouble identifying normal stationary obstacles in the car's path is a very bad sign. Sure, it's not outright impossible to solve this problem with a purely vision-based system -- but it IS much, much harder compared to fusing with other sensors.
Every time a new FSD beta releases, people post a bunch of videos that include Teslas trying to run into things, and a bunch of Tesla defenders hand wave away every incident as inconsequential, because it's inconvenient to their narrative.
Nailed it!
I suspect that Tesla made its life much more difficult by betting the self-driving farm on cameras. Which may have been a case of sales team overriding engineering objections.
This is just a guess as I have never worked for Tesla. But I did work on detecting and tracking various things and multi-modal sensing (e.g, camera+radar) is very powerful, as some things that are hard to detect, track or ID using one sensor type will be crystal clear after adding another sensor type.
> I think it would only take a couple of big issues to make the public distrust the tech.
The public, at least in the rich world, has become pretty bad at understanding risks. A single unlikely but highly dramatic event will affect preferences much stronger than a much more likely, but less media-genic thing with the same outcome (e.g., the death of a single person). This is not an easy hurdle to overcome, but it is possible if self-driving companies put more energy into comparing their safety against human drivers in a simple, clean, easy to understand form. My 2c.