I tend to agree. That's the point where the cost to them is actually reasonable in the scheme of things. In the mean time I'd just be happy with some new utilization of what we have...
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
HW2.5 capabilities
- Thread starter verygreen
- Start date
-
- Tags
- Autonomous Vehicles
d21mike
Active Member
Did you enable the option in your Settings? It is disabled by default. When I picked up my car on 9/28/17 I enabled it an other options. I was told it would not work until I drove 50 miles, then I was told it would not work till 100-150 miles and then I was told it was 60-80 miles. In my case it started working when my odometer hit 28 miles. I picked my car up at the Factory with about 4 miles.
I also got EAP and FSD.
I am very curious what @Bladerskb has to say for himself/themselves, assuming he/they don't disappear. Popcorn is out.
I am very curious what @Bladerskb has to say for himself/themselves, assuming he/they don't disappear. Popcorn is out.
Maybe @Bladerskb can comment here to keep this thread clear: Thread De-railers bitch about each other here.
Am I reading this correctly - please correct me if I am off by an order of magnitude: "Pegasus" runs at three hundred and twenty teraflops? EDIT: ok I just learned that an instruction is not the same as a floating point mathematical operation - so this is still a crazy 80 TFLOPS. WTF Skynet is on its way...
Yah, we knew for a long long time Xavier was coming at the end of this year with production the middle of next year. NVidia has said all along either 2 PX2 boards or 1 xavier board when it get's release and xavier has better thermals.
We have about as much chance of seeing Pegasus as we do seeing Perseus climbing down from the stadium walls with his magical sword and slicing off medusas head.
It's not nearly enough... only if you want to drive as badly as an inexperienced human driver, driving that road for the first time.
Also, if you want to drive with vision only, you'd need the system to understand all road markings, junction types, and - critically - all road signs. This assumes you have an AI attached to that vision system with NLP capable of human level contextual understanding too...
...which is a completely different ball game. Even I, with my human brain, struggle to understand some of the signage on the UK road network. Never mind some of these examples!
GPS is unreliable, it gives you ~10m accuracy on a good day. It's far less accurate when you're in a built up area, like a city. It's prone to losing lock, can take a few minutes to lock when you're leaving an underground car park, for example. It doesn't work in tunnels etc etc.
A HD map with localisation gives you ~3cm accuracy about the environment, and the position of the car within that environment. Thus, it gives you the drivable path. It doesn't care whether it's dark, or foggy, or rainy or whatever. It can tell you everything you could possibly want to know about your journey, all the road topology, before you leave your driveway. It's lightweight data, much lighter than the maps used for sat-nav (which weigh in at gigabyte proportions). The car can know everything you could possibly want to know about the drive, way more than a human could ever know, before you've even left your garage.
Good vision will give you the ability to avoid crashing into stuff, match traffic conditions, and deal with dynamically changing (templated) signs; like traffic lights. Even better if the vision knows where to look for those things, because of the map (helps avoid false positives). That's pretty much all vision needs to do, and we're pretty much there already - or nearly, anyway. It also gives you the foresight to warn the driver/passenger when the car *won't be able to drive itself any more* - making for a smooth and safe handover period.
This isn't just a fall back in case the camera fails - this is essential stuff. You can't have an autonomous car without a HD map.
The good news is that really, regardless of HW2 or HW2.5... DrivePX2 or Xavier... once you add the map and localisation into the mix and activate the surround cameras (plus add traffic light recognition), you've got a pretty functional system.
Anyway. these are worth checking out if you're not a map believer:
TomTom HD Map with RoadDNA | TomTom Automotive
End-to-end HD Mapping for Self-Driving Cars from NVIDIA Automotive
Whoever Owns the Maps Owns the Future of Self-Driving Cars
lunitiks
Cool James & Black Teacher
This is what I was talking about, earlier in this thread, when I was talking about Tesla not having HD maps. The hallmark of a true HD map is the localisation part -- it means you can locate yourself precisely in the map (within a few inches at least) without a GPS signal. In principle you can then drive without visible lane lines or street signs -- you can do better than humans in some ways. Trying to do this without LIDAR is handicapping yourself substantially. (Yeah, yeah, structure from motion blah blah... it's still way better with LIDAR.)
Well, it's possible to get to ~10kb per km using Mobileye's mapping techniques. Thats much, much smaller than regular, human readable maps. That's about 1 megabyte per 62 miles. I don't know if Tesla's maps would be that size, but it's possible.
At 10kb per km, I could drive the entire length of the UK in 14mb. This is easily streamable over LTE, 3G... even GSM at that rate.
Additionally, updates to the map would probably be delivered as a delta, and be specific to your route, so in reality they'd stream in at even smaller file sizes than that.
I don't know what Mobileye is doing, but that's not an HD map. They really claim to get 3cm localization without GPS using 10kB/km map data?
stopcrazypp
Well-Known Member
Again, you seem to be using a completely custom definition of HD map which is not in step with what everyone else is using. An HD map is simply a map with better accuracy than your typical GPS maps. It does not have to be 3cm.
Mobileye's accuracy goal is eventually to get to 10cm.
GM Exploring Mobileye Advanced Mapping With OnStar Data
Cirrus MS100D
Member
I'd wager to guess the small size would be due to it being more of a vector-based style/format, sort of like the difference between a high-res photo file and a vector image file. I doubt they'd need to store a wireframe of the entire 3D world like an Unreal Tournament map; you just need the stuff you care about and some method of identifying YOUR place on that map.
As was alluded to in other posts that highlighted some positions Tesla is seeking to fill, a dead-reckoning system should work in conjunction with all of this to help with positioning. If I know *exactly* where I am when I pull out of my driveway, my subsequent steering inputs (would slip angle need to be calculated in here?) + speed should have me pretty damned close to reality time and time again. Recalculate this position "periodically" (whatever that means) based on surroundings and horribly painted lane lines can be forgotten!
As was alluded to in other posts that highlighted some positions Tesla is seeking to fill, a dead-reckoning system should work in conjunction with all of this to help with positioning. If I know *exactly* where I am when I pull out of my driveway, my subsequent steering inputs (would slip angle need to be calculated in here?) + speed should have me pretty damned close to reality time and time again. Recalculate this position "periodically" (whatever that means) based on surroundings and horribly painted lane lines can be forgotten!
I thought this was interesting:
New GPS chips in 2018 smartphones will help you avoid that 10 mile detour after a missed turn
Superaccurate GPS Chips Coming to Smartphones in 2018
They claim new GPS chips in android phones would give 30cm accuracy.
More tech specs, great find @conman Superaccurate GPS Chips Coming to Smartphones in 2018
They are sampling now, so I guess iPhone 11
They are sampling now, so I guess iPhone 11
They're getting ~3-5 cm lateral localisation, and at least 10cm longitudinal, from vision only. 10cm is with sparse landmarks every 200m. In a more common, denser environment you get landmarks every 20m or so, which can improve accuracy further.
Add radar into the mix and you may get even better still, but really... it's already more than good enough to drive you safely through mind bogglingly complex junctions - ones you'd struggle with as a human - in pretty much any weather. No need for GPS, other than basic positioning to optimise the process.
These autonomous driving maps aren't "maps" like you think of them in the traditional sense. There's nothing to see. They're designed to be machine readable, rather than human readable, and they contain very specific information. The term "HD map" isn't yet well defined. It runs the gamut from a human-readable 'slightly better than normal' map to centimetre accurate laser-scanned maps from Google and HERE, to slightly less accurate guide-rail ADAS maps like Mapbox Drive. It's a wide term at the moment!
Tesla has been very, very quiet on the map front. I'm hopeful that they've been hard at work here, and are ready to deploy in some areas soon. When they do, our cars will seemingly get some new superpowers overnight - regardless of whether they're HW2 or HW2.5
IMO, "HD" maps are a crutch for propping up for poor software, and a nice plump revenue stream for those who generate & maintain them.
Example from Kyle Voigt's excellent blog post about autonomous driving in San Francisco:
Why testing self-driving cars in SF is challenging but necessary
Maps have their place (of course) but the value and utility of "HD" maps is being massively oversold.
Example from Kyle Voigt's excellent blog post about autonomous driving in San Francisco:
Why testing self-driving cars in SF is challenging but necessary
Maps have their place (of course) but the value and utility of "HD" maps is being massively oversold.
Similar threads
- Article
