I was kinda addressing the whole software package needed for self-driving rigs … but GPS map updating will make a good example … these roads we were on are at least 50 years old (judging by the size of the trees in that forest) … now, if we want to get to Moonfalls, we have to allow non-paved roads, and there’s where the problem begins … we turned off the main highway onto the 5847 road, and by every measure, the 949 road was a shortcut, less miles, less time, less everything … except we needed a high clearance pick-up … I just don’t see software companies driving all these miles of logging roads every year for inspections … Moonfalls is an about 75 foot tall, lush carpets of flora everywhere, National Park grade waterfall …
The Ford product will not have a steering wheel or pedals {Cite} … the only non-self-driving mode is “parked” …
So I’m driving up an incline on snow packed roads fishtailing a little … no big deal … wife puts her hand on my wrist “Honey, would you please stop fishtailing” … I check the kids in the back seat who were pure white in terror … passenger intervention …
Unfortunately Cecil’s column which started this thread merged all “driverless” cars into a single category. This has contributed to the confusion seen in the thread. Yes, IF there was only a single category of driverless car, you’d be worried how that might handle a wilderness National Forrest Service road in Moonfalls, Oregon.
But there is NOT a single category of driverless cars. The 2021 Ford you mentioned lacks user controls but it is ONLY designed for use in a well-surveyed controlled urban environment, not the wilderness of Oregon. Thus there is no conflict about how might a Class 4 driverless car or GPS maps handle an unpaved logging road.
That would be a theoretical issue – whenever Class 5 driverless cars are developed. There is currently no development plan or timeframe for these within the next 10 years.
There have already been demonstrations of car hacking. Now imagine several autonomous vehicles on the road at the same time. Now imagine even a few of them getting hacked at the same time.
I imagine that there have already been demonstrations of a 13 year old kid stealing the keys and sneaking out for a joyride while drinking tequila, talking on the phone and steering with their feet.
There are quite a bit more than a few human-driven cars on the road right this minute that are posing dangers you don’t even want to imagine.
I believe his point is that while there are singular bad drivers (although there seems to be more and more of them these days) a malicious inteuder could potentially hijack hundreds or thousands of vehicles causing mass destruction and interference. Security is definitiely a concern with autonomous vehicles, which is another facet that begs for a uniform standard of quality and verification testing.
Often, a passenger is also a qualified driver. And even if not, the passenger may occasionally see something the driver has not, and issue a warning that causes the driver to avoid a problem (IOW, to override what the driver had otherwise planned to do).
As a passenger, I’ve many times done this - and at least twice was able to avoid a serious problem by grabbing the steering wheel. The ability to do so could be a safety hazard, but also has considerable potential to enhance safety.
Not that we know of, although one hypothesis for the disappearence of Malaysian Airlines MH370 is malicious hacking of the navigational systems. Concerns about the potential for malicious hacking have been expressed by information security experts. Another thing to note, however, is that the firmware on aircraft control and navigation systems is uploaded by hardline by maintenance technicians, providing an “air gap” for security. The trend on consumer automobiles, however, has been moving to on-the-fly updates over cellular data connections, and using various wireless CAN systems, and has been aptly demonstrated the effort that car companies have put into securing thses systems is minimal if at all, often not even requiring any authentication much less encryption. The potential for malicious hacking exists can could be dire if a serious actor or even a particularly clever single vandal exercised an exploit.
It is certainly true that as of yet, car manufacturers and makers of systems like OnStar have not yet implemented much in the way of security. Thus, car hacking can be done and has been demonstrated.
It is hoped that as automated cars become more commercial, this concern is better addressed. Since there have been demonstrations of this susceptibility, we can hippie the industry is beginning to take this into account as the legal framework for financial accountability comes into place.
It is also possible the people at the top are sticking their heads in the sand.
The automotive industry as a whole has a noted lack of concern about “the legal framework for financial [liability] accountability” and the consequences of a wide scale malicious attack go far beyond the stakes of personal liability. There has long been the concern that the “Internet of Things”, which is not only network accessible personal and household monitoring devices but is increasingly becoming critical transportation, power, and communications infrastructure and devices, is incredibly vulnerable to a concerted attack as evidenced by recent cyberattacks on the Ukraine and other nations. Disabling, or worse, causing vehicles to drive out of control during high traffic times could not only pose personal hazards but would be economically and socially disruptive as well.
Securing network-connected systems against malicious outsider attack, while not trivial, is feasible but requires building in a cybersecurity infrastructure from the bones out. Adding ad hoc security measures to an existing system not designed for them–especially bare bones embedded controller firmware designed for maximum compactness and processor efficiency rather than enabling authentication of commands and sensor data is not really feasible, and having industry-accepted standards for the design and testing of cybersecurity measures and features ensures a uniform standard will be applied regardless of the design cost.
I read “even a few of them” as meaning significantly less than hundreds or thousands.
But sure, security is a concern even in the case of just a few. My point is just that we aren’t terrified that there are already thousands or millions of extremely unsafe drivers operating a car right now with no hacking required to make them dangerous.
Even assuming thousands of autonomous cars were able to be hacked giving full control of the vehicles to the hacker, which is an improbable scenario given even a modicum of security that they will hopefully implement by the time they are mass produced, in balance there will also be thousands or millions of dangerous drivers happily letting their cars do the driving for them. Even in a worst case hack scenario overall the roads would probably be safer. And I think the worst case is an unlikely one.
Every day we get thousands of demonstrations of people drinking alcohol and then attempting to drive a car. In 2015 35,000 people were killed in the US in car accidents. While hacking is a concern it’s down the list.
There is no question that an SAE Level 5 autonomous vehicle will be more attentive and capable of making faster and generally better decisions than even the best professional driver, and that this will translate into orders of magnitude of reductions in the vast majority of accidents, even those with external causes such as wildlife impact, inclement conditions, et cetera, and with intertraffic networking (communication between autonomous vehicles in proximity) they will be able to operate more efficiently and safely. This does not, however, offset the potential for vulnerability of networked systems. That potential is already realized, and is only limited by the relatively small amount of control that current automobiles have; they can basically disable certain features or systems but in most cases the driver can override the vehicle and pull to the side of the road (although drive-by-wire systems in the throttle and brake have already compromised that control on many newer vehicles). The assumption that auto manufacturers will implement “a modicum of security” is a dangerous one; over and over we’ve seen how major car manufacturers will accept known problems and compromise on safety to save the expense of a few dollars, which can translate into tens of millions of dollars of profit over the production lifetime of a vehicle. Spending money on features that do not add any visible value to the vehicle in the eyes of the customer is low on any manufacturer’s list of priorities, e.g. the long period between the introduction of seat belts, and later safety features such as collapsable steering columns, crumple zones, and anti-lock breaks, and the near universality of these features even in mid-range vehicles.
The problem with vulnerabilities on autonomous vehicles which control all aspects of vehicle operation aren’t limited to a few obnoxious hackers; a single exploit could pose a critical vulnerability in an entire model, or even multiple models if (as is likely) the system is duplicated across models, potentially exposing hundreds of thousands or millions of vehicles to malicious override by, say, an unfriendly foreign power. If your autonomous car can’t be trusted take you to work or causes massive congestion it can have a severe fiscal impact implemented on a wide scale.
This is not a novel concept; information security experts have been warning about the danger of this kind of exploit on various networked systems for over two decades, and it has taken this long for it to come into the awareness of the general public; it is analogous to airliner security and counterterrorism experts warning about the possibility of hijackers using airliners as weapons of attack since the 'Eighties, and nobody did anything about it until nineteen fanatics crashed airliners into the Pentagon and the World Trade Center. Given the expectation of handing over our entire over-the-road transportation infrastructure to autonomous systems it is entirely sensible to pay head to warnings about securing those systems from intrusion and malicious actors, and not just in a “modicum” of effort or degree.
Good idea! Nobody’s ever managed to hack a psychedelic-paint VW minibus; let’s base the industry security standard on those!
I’m not so sure about that. You only need security where the car’s computers connect to the broader network. It could be as simple as verifying that any incoming data is signed with the manufacturer’s correct digital signature.
I"m pretty sure the first widespread use of autonomous driving will be on highways. The first industry to widely adopt it will be long-haul trucking.
At first, there will need to be a human driver on board, but that person will only take the wheel when the truck needs to leave the freeway.
Some people will be worried about such large vehicles being computer-controlled, but it will have big advantages. Right now, transport is either slowed down by the drivers needing sufficient rest, or made dangerous by drivers that haven’t gotten sufficient rest.
Plus the cost of transport will go down once the trucks can function without human control. Lot of drivers will be out of work, though.
My prediction: we won’t have self-driving cars before 2030, in the sense of ordinary folks being able to go down to their car dealer, and buy a car that drives itself from one place to another without the rider having to pay attention to what’s going on with the car in between. No earlier, and maybe a good deal later.
The task of recognition of thousands of ordinary situations we encounter in local traffic and handle with minimal thought - that task seems to still be way beyond what autonomous vehicles can manage. Well before then, self-driving trucks may be able to handle going from one depot directly off an interstate highway, to another depot right off the interstate 300 miles away. But that will be feasible - if it is - by the fact that that excludes all your local-driving problems.
That task is already within reach for computers. It’s still well beyond what humans can manage, though.
And I’ll bet that another early adopter will be school buses, just because the supply and demand is currently so skewed. Every single school district in the state of Ohio, and I imagine the majority of them elsewhere, has openings for bus drivers. You’ll still need an adult on board to supervise the kids, of course, but that adult won’t need a CDL, making the supply much greater, and will be able to devote full attention to the kids, greatly improving safety. School buses also drive a consistent route every day, which will make it a lot easier for the computer. I suppose you might have a handful of human drivers who “train” the buses on their routes a few days before school starts, but that’ll still be a lot easier than needing a human for every route, every day.
