Found this at Axios!
It takes a variety of overlapping sensors working together for self-driving cars to accurately perceive the world around them.
Why it matters: Cars that drive themselves — whether in limited highway settings or in geofenced urban areas, like robotaxis — require superhuman vision, along with sophisticated prediction and decision-making capabilities.
Full article and graphic HERE.
This is probably the best ‘simple’ graphic I’ve seen that shows the various overlaps required for ‘self-driving’ cars.
While they don’t go into much depth, there is ALSO a requirement for a very fast very smart computer to process all those inputs and make ‘decisions’ following whatever trees are applicable for any particular system, and which systems are the priorities!
Autonomous driving will not be complete until it can take directions from wives.
Interesting, short though not very in-depth. It far underscores the complexity of the issue of autonomous driving vehicles.
One thing they are amiss on is the resolution of automotive radar. It is at a frequency of 76 GHz (76 billion cycles per second) that translates to about 4 mm wavelength. Thirty years ago I saw some pictures from radar at those frequencies; they looked like black and white photos of the scenes that were scanned. One could identify vehicles in those pictures.
I think Ed has it right.
How do they plan to mitigate RFI? How will each car know it’s own signal return? And a road full of these things may require shielding like on a stealth fighter canopy for the hooman to exist in the thing.
Ranges are much shorter than anything that aircraft might need to deal with, so the power probably is drastically reduced. That may mean that the risk to humans of cancer, etc., could be minimal.
The budgets for the radars on these are very unlikely to be anywhere near what can be spent on military radars. Nor is the power available likely to be very high.
Which doesn’t make the choice, or anything else involved wise, or sure to be done correctly…
I hate driving my wife’s 2017 RAV-4 for more than just short trips. For longer rides, I take the time to disable all the annoying nannies (lane departure, predictive braking, etc.), but especially the adaptive cruise control. The radar picks up all kinds of crap I would never disengage for, so I disable it when I turn it on. It will “see” a tractor trailer in another lane on a curve, an aggressive driving jackass that cuts across the front of my vehicle while traversing several lanes at high speed, etc. If you hold the cruise control power button in for 1 1/2 seconds or so it disables the radar and operates like standard cruise control.
I’m on my 3rd generation of Fords with adaptive cruise control and I haven’t seen the false positive problem. I do adjust my trailing distance depending on the traffic and speed. On the open road max trail means I never get close enough to the unit in front to pass easily, but min trail means getting a bit too close and having to break harder. The Ford system can control my speed all the way to zero and start back up again if the pause is fairly short. From my testing the Toyota system wasn’t able to do either (but the Toyota rep had no clue on how his system worked or it’s abilities).
I once drove a Corolla with all that crap on the Turnpike. Just cruising alone visiting with the eldest child, Looked down to see that I was 10 mph under the limit because of the adaptive CC was seeing a slower car ahead of us. It could at least be set up to give notice that it is slowing you down.
Ed- LOL
Bill- MM wave radar has ‘excellent’ resolution, no question.
STxAR- Ah, the $64 question…
Glypto- Concur!
Gregg- One would think, but then it’s a computer, so it knows what it is supposed to do, you don’t… sigh
Tesla’s experience with AI is one of the reasons that Musk is backing off the AI wagon, so to speak.
The folks in Eurostan who found an unattended “smart” car at a mountain road turnout, and “trapped” it permanently, by simulating a “no crossing” line around it with a canister of table salt, told you everything you need to know about the viability of driverless cars.
Imagine what happens when people start randomly spray-painting double-yellow lines perpendicular to road lanes every 50 yards or so, and putting up random STOP and WRONG WAY signs in the middle of city avenues or endless deserted highways, or putting RIGHT TURN ONLY signs pointing over the edges of cliff-hugging canyon roads.
Hilarity ensues.
Game over.
From my observations the AAC systems rules are to 1: pace the car in front OR 2: go the set speed limit, and it is up to the driver to pay attention on what it is doing. Yes, I’ve spent time behind a semi because of it but I’m happy supervising the truck on a 400 mile drive vs. actually driving the whole way.
Beans- Exactly!
Aesop- Sigh…and idjits WILL just to see what ‘happens’…
Rick- Interesting perspective, and I see your point.
A bit of background: I used to live in Orange County CA but had to go to LA and Points North on a regular basis, so I was driving in Stop-and-go freeway traffic for miles and hours. Once I knew I could trust my Fusion to handle pacing the car in front to a dead stop then starting back up again when traffic moved it reduced the stress of getting to and from a call dramatically. Not an autopilot by any means, just keeping distance. I custom-ordered a hybrid Explorer with the next generation ACC system with lane-keeping and survived being the only driver on a road trip from CA to Minnesota and back with my daughter two years ago.
You have to pay attention but the mental workload is MUCH lower. Your P-3 pilots used their autopilot for the same reason.
Greggbc: Adaptive cruise control slows to set the number of car lengths to maintain.
Autonomous cars will not happen, at leadt, until all cars have vehicle-to-vehicle communications. In a 2-D world there are too many sources of interference for sensors only automation.
Burying a wire into the payment in the middle of each lane, especially with some ability to convey location info, is another possibility to aide in autopiloting.m
I do not want any of them bells and whistles at this stage.
Automatic transmission has ruined the proper training of drivers.
I’m currently working with a semi-robotic land-based system. Sensing the environment and developing a course is very complicated. Radar and LiDAR really only see in 2D, so have to fuse two or more sensors inputs, recognize and identify items in their view, then navigate. Not easy as others have said. Add that different wavelength of sensors provide more or less fidelity with the attendant problem of sensor range and overlap. See “range gate” and “range gate stealing” as examples. The sensors operate great in a friendly environment. What happens if there is a little ECM in the area. UH-1 “disco ball” or something 40 years improved to really screw with your sensors.
I’ll stick with my older technologies, thankyouverymuch.
Rick- You do have a good point!
Jaime- Agreed!
WN- Thanks for chiming in! Yes, ECM (or other systems doing the same thing) WILL screw the pooch big time, as we know…
Article writer missed some bits.
Issue with radar isn’t necessarily best worded ‘accuracy and resolution’.
So, the two radar types, the ultrasound echolocation, and the lidar, have somewhat similar operating method. They are active sensors that use reflections. This is less a problem with lidar, but for the others you get reflections from everything, and need to spend some efforts filtering, and resolving the actual targets you need to keep distance from.
Visual cameras work like we are used to thinking of sight working. But, maybe a bit worse than human eyes. Lidar are in light frequencies, so same optical properties, but the data is not recorded the same. IR is in between microwaves and visual light, so IR cameras can be vision like, but may not necessarily be vision like.
There is a lot more details that can be given on these approaches.
Though, you do not need the sensor coverage or the the processing or controls to make a self driving car if you can find enough engineers who care nothing for liability and who care nothing for human welfare.