Comment by walrus01 | Hacker News Reader

Comment by walrus01 | Hacker News Reader

Parent Full thread

This seems like a setup that while a logical architecture decision, would be reliant upon really rock solid mobile broadband on the routes driven by its trucks. Like, seven nines reliable. I saw videos of their test setup with a driver at a desk workstation/"cockpit" which was driving a remote truck.

Imagine if a human needed to make a decision with less than a few seconds of reaction time, at freeway speeds, and the LTE network where the truck happened to be at that moment was having a hiccup.

It's much worse than a cargo drone that might have a 12-14 satellite GPS+GLONASS lock, that could be programmed to stop and hover in place, or retrace its flight path breadcrumb style back to base in the event of brief loss of WAN network connectivity. It's something with possibly 20,000 kilograms of mass moving at 70 mph with live humans in traffic all around it, and a lot of inertia.

It's an external technical problem that the company itself can't solve, because they're not in control of LTE/5G network operators, they don't make LTE radios/modems/antennas, etc. The only places that I'm aware of using fully remote driven trucks are giant open pit mines in western Australia, where the mining company fully owns and operates in-house the network linking the trucks.

I personally know very little about self driving technology, but working in network engineering, the idea of relying upon the normal publicly-accessible nationwide LTE network operators for something that could be a life-critical safety function really scares me. Even with a max-priority service like Firstnet on ATT it would be sketchy.

The trucks could have AI that worked most of the time. Not well enough to rely on for normal operations, but good enough for a fallback in case of losing connectivity.

Some made-up numbers to explain my point: Consider a situation where the remote operator will make the right response 99.9% of the time, failing 0.1%, and the AI is 20 times worse, making an error 2% of the time. The actual error rate will be (0.1% * connection reliability) + (2% * (1 - connection reliability)). For 3-nines reliability, which is pretty poor, the error rate will be about 0.102%. For 4-nines it will be 0.1002%.

Those sorts of figures would effectively pay for themselves in terms of safety, once you factor in the benefits for monitoring and work conditions for a desk-operator vs a physically present driver, and they don't require either a particularly reliable connection or a particularly good backup AI.

Ah yes, the mythical "AI" that has yet to be invented.

Let's just sprinkle in some AI and it'll work!

The exact cases where human reaction times fail are already taken care of by local subsumptive AI† in cars. For example, collision avoidance is “AI”; lane-keeping and adaptive cruise-control are “AI”; etc. For that matter, an automatic transmission is “AI” in the sense of abstracting away a clutch and gears under a prediction of when you’d want them used.

† ...which doesn’t actually imply anything to do with Machine Learning. Your computer downclocking when it overheats rather than catching on fire, is an example of “subsumptive-architecture artificial intelligence”, even though it’s a hand-coded algorithm doing the subsumption. The “artificial intelligence” in this case refers to the emergent property of the layered subsumptive processes, not to the processes themselves.

Don't forget anti-lock brakes.

I think their goal was to have highway driving completely handled by the self driving systems with maybe call outs if it gets into a situation it can't figure out (weird construction zone or accident it needs to move around). For covering the gaps existing technologies might even be enough, a combination of auto breaking, adaptive cruise control and lane keeping would be a pretty solid gap fill.

There's several things that could be done in the city if it experiences drop outs. A creep mode where it moves quite slowly till the network improves would let the lower bandwidth or latency connection still be used safely.

Personally I always thought the answer would be self driving on highways to cargo hubs for delivery by human drivers. Or maybe self driving trucks with sleepers so if it got into a confusing situation it could stop/pull over and wake the driver to deal with it and allow the driver to sleep while the truck keeps moving.

>maybe call outs if it gets into a situation it can't figure out

How would it predict that these situations are coming with enough time to hand off to a human driver? I think it would need some form of crowd sourced or externally gathered data. The truck itself couldn't gather this in time.

>A creep mode where it moves quite slowly till the network improves would let the lower bandwidth or latency connection still be used safely.

Sounds like a traffic nightmare.

>Personally I always thought the answer would be self driving on highways to cargo hubs for delivery by human drivers

I do agree with this. I firmly believe that full self driving will probably never be possible for anything except for Interstate/main highways that are well maintained and very standardized in their construction, markings, etc.

There are source out there for accidents, traffic information and construction where the system could pretty reliably predict when it's going to encounter an odd traffic pattern that would be better handled by a human than the automated systems. In cases where there's a decision needed extremely quickly we're generally talking about an accident or near accident that relatively simple existing assistance systems like autobraking and obstacle avoidance would handle.

Trucks are already really slow compared to traffic inside cities and in the rare cases where they're in this mode on a highway they're probably already in a slow traffic area because they're dealing with construction or accident areas.

Reply on news.ycombinator.com