I certainly see this coming to pass in conjunction with various levels of autonomous vehicle control.
On highways vehicles equipped with both V2V and adaptive cruise control could form ad hoc virtual trains.
As for cows and geese … those are no harder for automotive AI than they are for people. Probably better at avoiding the deer. For water it is likely that such vehicles will be better able to assess if it is safe to proceed than people are, by virtue of sensors other than sight alone, by virtue of potentially better information about the exact stretch of road on its topography, and in the case of a V2V world potentially by information communicated by multiple vehicles that have recently traversed the section.
If you think about it, ranchers and police and military and firefighters will all have special manual drive vehicles (with the usual highway autopilot if the vehicle is able to go on public roads) long after it’s basically illegal for anyone else. I’m kind of imagining that in the far future, people will be excited to learn to drive for real when they are in training for such programs, seeing it as a real perk and an exciting task, not the chore we see it as.
Off-road autonomous vehicles are probably more advanced than in-town driving mainly because of the long running DARPA Grand Challenge. There’s a lot of algorithms for detecting and navigating around bodies of water, from puddle avoidance to river avoidance. Whether there’s a specific algorithm for dealing with washed out roads, I’m not sure but it wouldn’t surprise me.
Cows, geese and other living things are becoming easier and easier to detect as the algorithms becomes more advanced. I’ve seen algorithms tested on pictures of horses, tigers, cows, ducks, etc. There’s actually a student’s conference poster that they tested on horses right outside my lab.
Thanks for sharing that. I hadn’t seen anything about it and as a tech solution it helps autonomous systems while providing for enhanced warning systems for manual control. It’s especially useful for situations where both are on the road.
Do you have a reason to think they’d be easier for the system to avoid than the other fleshy speed bumps? Based on a lot of experience driving in northern Michigan they are one of the hardest things to react effectively too.
An automous driving system would stop before running into a herd of cows or flock of sheep. It would recognize the that “a field…turned into a car park” has standard features of a drivable area and respond accordingly and likely send navigation updates to a distributed geographic information system accessible (GIS) by other vehicles to alert them to a change. If it detected standing water on the road it would slow and look in its GIS system to see if flooding is a concern in the area, and proceed slowly, reversing if it detects water too deep to traverse.
It seems many of the complaints about hypothetical future autonomous driving systems are based on some kind assumption that they will operate in some fashion with no margin for error, blindly following some kind of pre-programmed instructions, with no ability to cope with any hazards not explicitly defined on a map and without the necessary information to determine if a road hazard exists. Current driver assistance systems on higher end cars are already capable of making these kinds of distinctions separate from any kind of preprogrammed navigation, and they are comparatively primitive relative to the requirements for an autonomous system. Autonomous systems are going to have to be designed with heuristic capability, able to identify new hazards and apply a set of general rules to develop ways of dealing with them…just as a human driver does. This bears about as much resemblence to your “Flappy Bird” app that crashes five times a day on your iPad after you updated the operating system as bloodletting does to laparoscopic surgery.
And regardless of how you feel about the reliability or viability of these systems, they are coming, and probably faster than most people expect once certain thresholds of capability are achieved. I doubt we’ll see large scale adoption of autonomous driving technology in five years, but I’ll also be surprised if most people are still driving themselves in twenty-five years, and the safety and reliability of these vehicles in driving in all kinds of conditions will be an order of magnitude better in terms of vehicular accidents and pedestrian impacts than current meat-based driving systems while virtually eliminating traffic jams, intentional obstruction due to road rage or inattention, and generally contributing to safer, cheaper, and more efficient personal transportation. So, you can live in some kind of fear about self-driving vehicles, or you can be informed about the methods applied to ensure that they are safe and encourage governments and consumer organizations to demand rigorous testing methods and transparency in software protocols to ensure that manufacturers are not cutting corners to get immature technology to market soonest.
These are the kind of smart, safe decisions human drivers make. “Hey, a giant cloud of opaque and potentially toxic black smoke. Let’s drive into it and see what happens! I don’t want to be late for Zuzu’s dance recital, after all!”
This too!! We could add falling rock zones too. it might make the DC play ballads from my Zepplin tunes instead of swerving properly.
And to further clarify the type of fog that keeps coming up, I’m talking about sitting on a freeway, 0 mph, no cars moving, can only see the tail lights in front of me, can’t read the plate 5 feet away and 1/2 a mile back visibility was fine. Testing for these sensors will be tough as hell. And they’ll be on every DC?
Those conditions, the safe speed is about 5 mph. Even current machine vision systems are going to notice the lack of data from anything farther away than a few feet, and the appropriate speed is calculated by a fairly simple equation.
Radars do see through fog of that type just fine, and so does ultrasonic. I don’t know if IR lidar can or not. So the car might still maintain a moderate amount of speed, depending on how confident the programmers were about those sensor channels.
Nobody is saying you cannot come up with a set of conditions that won’t crash the car. It doesn’t need to be absolutely perfect, just on average do about 10 times better than the average human driver. 1/10 the death rate is an enormous improvement and we shouldn’t be against these cars because they will kill 10 people for every 90 people who get to live.
One thing I’m curious about. How are speed limits, particularly on Interstates, going to be handled?
I drive a lot on I5, from Seattle to about 35 miles north. The posted limit on this stretch is set at 60 mph, while the actual speed driven by most of the drivers, when traffic is not bumper to bumper, is from 68 to 72 mph. And I know for a fact (having tested this more than once) that the State Patrol doesn’t stop anyone doing less than 70 mph if the driving conditions are safe and traffic is generally flowing at that speed.
Are driverless cars going to be programmed to not exceed 60 mph? Or are speed limits going to be raised to reflect reality?
Easier for the system to avoid deer and other fleshy speed bumps than it is for human drivers to do, for sure.
A variety of sensors that operate with a 360 field of vision including off the sides of the roads and in ranges inclusive of 300m including through fog and on unlit areas at night (visible light cameras, radar, lidar, ultrasound, and infrared lights to illuminate for infrared sensors, in various combinations depending on the vehicle).
100% vigilance 100% of the time.
Faster reaction times.
Not sure about speed limits but Google’s general approach is to collect data about how drivers drive in the real world and have the cars learn for that. The rule the when cars come to a 4-way stop at the same time the car on the right goes first is disregarded in preference for what the systems have learned real drivers do … whatever that is. Not sure but it would not surprise if the programming understands that being much slower than flow of traffic is relatively unsafe even if the alternative is going over the speed limit some.
“Sooner than you think!” is a real nothing of an answer. So I’ve adopted it for all questions of a temporal nature. When’s that report going to be ready? Sooner than you think!
I do detect a bit of American-Centrism in your post :). Sure driverless cars might be dandy in spread out American suburbs. But in megacities with tens of miliions of people on the road?
Autonomous driving systems will certainly be deployed first in developed nations such as the United States and Canada, and the industrial states of Europe and Asia where the benefits to safety and efficiency will justify the deployment and regulatory costs and traffic laws are (somewhat) consistently applied. As experience with the variable nature of real world interactions grows and the cost of integrating such systems drops, it will no doubt be adopted by other nations with more complex inteactions between automobiles and pedestrians. Again, autonomously controlled vehicles are not going to be racing at high speed through pedestrian-laden areas with wanton disregard, and the ability of an autonomous system with much wider sensory capbility than a human driver will reduce the substantial number of accidents that occur in those nations. How long will that take? It depends on the amount of effort and impetus, but after certain thresholds are achieved (such as systems becoming suffiicnetly heuristic that human intervention in designating the ‘best’ solution is no longer required) the reliabliity of these systems will grow in geometric fashion. The very fact that failures of autonomous systems will be scruitinized while regular faliures of human drivers are accepted as “business as usual” will drive the reduction of acceptable accidents to a minimum, and once it becomes apparent that autonomous systems are orders of magnitude more capable than human drivers, liability and insurance costs will drive mass adoption, notwithstanding the conveniences addressed above of being able to send a vehicle to do basic transportation chores without having to accompany it.
You know, whatever algorithms, software middleware, and specialized chips they develop for this should be broadly applicable to other robotics. The overall algorithm - the future state modeling, all the hundreds of methods and tricks for analyzing sensor data, and so on should make other automated devices work better than they ever have. More general purpose factory assembly could be made that just need to know what the end result is (from an assembled CAD model) and approximately what the subcomponents look like (that image could be a mere rendering made from the same CAD model).
Food preparation robots. Robots that stock store shelves or pack boxes. Robots that take inventory. Robots that conduct inspections. The list goes on and on and on. Probably far longer than any of us realize.
I detect a bit of human intelligence bias in AK84’s post.
The Chinese and Indian markets are not one and the same but both represent large opportunities for autonomous and connected vehicle technology. China has recently released a “roadmap” to try to “fast-track” adoption of the technology. Didi (the big ride-hailing company of China) is going to use some of Apple’s $1 billion investment in them to develop and/or deploy self-driving technology. Tata (of India) is going to try to be a player in the space as well.
The fact is that it driving better than most drivers drive in those countries is a low bar.
And there is less cultural baggage of cars than in America where cars and driving are enmeshed as part of the American cultural mythos. The premise of the op is well founded: no matter how good autonomous vehicles and no matter how well documented those advantages become there will be a significant resistance in many quarters to them. Culturally many Americans like to think of themselves as individuals in control of themselves and explicitly giving up control of what they are used to having control of will be hard to do. Eastern cultures are not as highly individualistic.
Highly congested traffic areas are an area where autonomous connected vehicles could offer huge advantages, allowing vehicles to pack themselves together closer while improving safety.
The difficulty in those markets may be cost more than anything else.
And often people do have a lot of experience with systems much more reliable than the ones they have at home - but don’t think of them as “computers”. I can’t remember the last time I had a problem with a supermarket’s POS that wasn’t a matter of either human training (something came up with the cashier didn’t know how to do) or human stupid decisions (some genius decided that items which have individual barcodes but are often sold in a pack with a different barcode shouldn’t have pack-codes entered into the computer). But you pretty much need to be someone who knows POS doesn’t only mean “piece of shit” to understand that thing is a computer.
Conversation on the subject, this past Saturday:
friend: - Thing is, I like driving!
me: - Oh, I like driving, too! But I can think of a lot of people that I’d like it if they didn’t drive!
friend: - … got that one right, I guess. I still wanna be able to drive, damnit.