These things are using neural networks intensively, which is far from algorithm-based and is far more like how we process information. Sure, there’s still plenty of algorithms for path-planning, sensor fusion, etc. but the really hard stuff is all NN.
Consider that freeway driving has far fewer use cases than urban driving and you see why trucks will be the first to use the technology.
I’ve been writing software for 35 years and work with many OEM’s creating autonomous vehicles. I guarantee you it is coming. For passenger vehicles it is a different story, you need it to be far less expensive, along with the ability to handle far more complex scenarios - so besides from having a redundant system for fail-operational backup, the human will be told to take over, as they’re the backup to the backup. For commercial vehicles they don’t need to be cheap, you can have more than one backup with everything using high horsepower compute.
There’s still challenges but anyone saying “can’t be done” doesn’t understand the current state-of-the-art, it’s incredibly advanced technology. They’re on the roads today, still with a human but as the miles rack up and the error rates drop below human averages the truck driver will go away - NHTSA will allow it as they’ve already stated.
They should never be autonomous. Why? The condition of the roads themselves, The term “rough service” is a huge understatement. Trucks are not “solid state”. There would be no profit in this trend because of downtime. Trucks do not make money when they are sitting on the side of the road with a load of perishables going bad. Have you ever physically worked on trucks in either fleet or retail maintenance and repair?
I have done both all my life, tens of thousands of them over the last 45 years. 95% of repairs needing done on heavy trucks now days are failed sensors, connectors, and wiring issues from extreme vibrations along with heating & cooling expansion & contraction. It would be impossible to maintain closed working circuits for these systems no matter how good the computer and software are. The power and signal delivery system for electronics are very unreliable with the minimal computerization they already have now. Intermittent momentary open circuits are a huge primary and perpetual problem with these systems.
Once had a whole fleet of 350 brand new trucks go down under load at about 10k like clockwork. And not just ours... Everyone who had one made that year. There were not enough sensors stocked and available for all these new trucks going down across the country, they could not make the replacements fast enough. The manufacturer was shipping sensors from/to all over the country to these broken trucks. It was perpetual for almost a year because they absolutely refused to redesign the sensor and kept replacing with the same bad sensors.
That one sensor cost our company millions and millions in late loads refused from the customers and delivery price renegotiation. The sensor? A simple coolant level sensor that was programmed to shut the truck down if they fail. Coolant level fine, but because of no reading from this unnecessary “go/no go” sensor all these trucks were rendered unusable by the computer. In the mean time neither Peterbilt or Cummins would step up and take responsibility for the sensor and warranty it. Both denied ownership of the part because the “cab computer” belonged to Peterbilt, and the “engine computer” belonged to Cummins. This sensor was one of those “marriage parts” of the two systems so they ended up blaming each other for the problem. Cummins engineers designed it, but Paccar/Peterbilt had it physically manufactured.
There is already a very fine line between profit and loss in that industry which absolutely depends on reliability and 100% up time. The less to break, the more dependable they are going to be. The electronic complication in this application and industry is already going far beyond that point and becoming grossly unproductive. Let alone the human nature factor that is never considered like the conflict between Peterbilt and Cummins over a simple “unneeded” sensor... A simple coolant level “sight glass” ALWAYS works and requires absolutely no wiring, power, or signal.
No, it has already gone beyond all common sense and logic, it is not needed, and should never happen. Let alone consider allowing electronics to actually physically control the vehicle... No matter how dependable it is claimed to be, it will still always be “driver error” anyhow. The engineers and makers will never take the blame for their own screw ups, so just let drivers be drivers. Just the perpetual open circuits alone in this system application will be far far more dangerous, less reliable, and less profitable.