Interesting research paper on the current short comings of AI in automated vehicles:
http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u36/reality%20check%20final_compressed.pdf
Interesting research paper on the current short comings of AI in automated vehicles:
http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u36/reality%20check%20final_compressed.pdf
You know, you can’t drink all day if you don’t start before noon.
I thought of a potential options for self driving cars. Think about recording a macro in Excel. If there was a common drive that you did (work, dropping kids off at school, etc) design a record feature where the driver drives the route and then the car can repeat the route and all it has to do is stay in the lane you chose and manage the accelerating and stopping.
Recording macros is not the best. It is inefficient to say the least. It includes scrolling and window changing and even simple codes like borders it does actions on all the borders that nothing happens too.
I imagine driving a car if you recorded a macro. Went up to 50 mph, 51 mph, 49 mph, 50 mph, started turning left, stayed in lane, signaled, stop signaling, switched lanes, etc.
It might be a mess without a clean-up.
Surely the “macro” would be of the form, “start here, end there, follow this path to get there, figure out the details yourself, depending on circumstances.”
The path would include appropriate stops. But surely it wouldn’t be so precise as to tell the car which lane to drive in, and certainly not how fast to drive.
That’s kind of what I was thinking. Help the AI learn the route instead of it trying to figure everything out. Like maybe it runs it’s program simultaneously and looks at all the things they driver did differently and recalibrates from there.
I have a feeling that the AI already knows all of the possible routes before you even put your car in gear and get on the street.
The problem becomes one of selection . . . which route to take. Most optimizations algorithm will focus on (estimated) time and selects the route with the min. time.
Additional constraints can be placed (e.g., “No Tolls”). You could also optimize distance (shortest distance isn’t always the fastest when speeds are very different for each route).
Personally, I’d like to see an option of “fewest road changes” for route selection.
Yeah google and apple maps already have most of those selections, in addition to things you don’t know (projected traffic etc.).
I doubt the AI would want your help to learn the route.
Interesting vid about unique experiences in a Waymo. This guy (college student in AZ) has taken 160+ rides and documents a lot on his website. Briefly shows the Waymo handling a crossing guard and a cop directing traffic among other novel situations. It’s still only a small section of AZ suburbs where weather shouldn’t be much of an issue, but cool to see. He was quoted in another article that unprotected lefts have seen a huge improvement over his time riding. They used to be pretty hesitant, but now it calculates the opening and just goes.
Here are three of his observations on how the service has changed since its public rollout in October 2020:
- Waymo “used to wait and be a lot more hesitant” with unprotected left turns, Ricks said, which are “pretty much flawless now. …If there’s a gap, and [the vehicle] knows it’s safe and can calculate everything, it will just go for it.”
- Navigating busy parking lots used to mean “lots of whiplash, braking, and in some cases, disengagements. …Now, there’s a lot more smoothing out and it feels a lot more confident.”
- The in-vehicle experience has stayed fairly consistent, with the exception of a plastic visor installed during Covid.
twig93 more like twig75 amirite?
Yeah, 93 is when I graduated from high school. I was born in 75.
In my experience, Waze seems to find some arbitrary number of possible routes (like 3 or 5 maybe) and then pick from those.
I’ve had it pick a really dumb route so I go a better way (like I know how to get to that part of town but not the specific address) and when it gets to the point that it re-routes based on what I’m actually doing it knocks 6 or 8 minutes off the arrival time.
No tolls or anything, set to quickest time, not shortest route.
So it seems like it’s not even considering my preferred route until I force the route on it then it’s like “oh… this is better”.
Also, it’s algorithm definitely doesn’t penalize turns enough… especially left turns.
Going from my house to Costco you can take a semi-main road (35 mph) 1.1 miles, turn left, and then take another semi-main road (40 mph) 3 miles… OR you can zig-zag through residential streets, turning every 0.1-0.2 miles where the speed limit is 25 the whole way and it’s like 0.2 miles shorter.
Even set to “quickest” it will pick that route every time, which is insane.
Google maps puts a higher priority on staying on large roads and avoiding turns than does Waze. You have options, even among the Google products.
Google maps have always been fairly accurate down to the t. I believe their engineers are smarter than me, so I’ll believe them.
If you’ve studied non Euclidean geometry you’d know that the quickest route is not at all obvious.
Not to mention they have information that I don’t, aka traffic status and projections.
That goes without saying
But I think the bigger issue is that Google likely have a patent filed (and possibly approved) that prevents other navigation apps from using some of the algorithms (unless they pay a substantial fee to the Googles).
very applicable for air travel; less applicable to a local linear space.
land travel is anything but linear
Is there an option for which route takes you past the greatest number of Hooters restaurants? Asking for a friend.
lol
I do believe that there an “Add a stop” option available.