Dana
Final Approach
Or the data from a FSD crash is more valuable than the cost of the insurance payout.
Crash at Reagan National Airport, DC. Small aircraft down in the Potomac.
- Thread starter FlightmechH3
- Start date
Warmi
Line Up and Wait
Possibly, but in any case they are confident in their software enough to make that a reasonable proposition …
THill182
Filing Flight Plan
- Joined
- Feb 11, 2024
- Messages
- 1
- Display Name
Display name:
TulsaPilotM20R
Excellent post; bottom line is that statistically/mathematically speaking rare events happen; and also tend to cluster, e.g., if you throw a hand-full of peas over your shoulder at a large group of people (e.g., like a classroom of students) than most and often all people won't be hit by a single pea; but sometimes a person will be hit by one, and sometimes two or more peas. That does not mean there is anything wrong with the person who got hit! Humans have the propensity to see patterns in this, i.e., blame the person (e.g., "accident prone"...) when it is a mere statistical artifact of rare (Poisson-distributed) events.
So I fear the "causes" that will be discovered finally will not really be root causes, but hopefully will lead to much wider safety margins to make such terrible disasters even more unlikely. I am not a controller but a formerly very active private pilot for almost 20 years who has flow into very active air spaces: So greater separation, better staffing and training for greater risk aversion (vector the helicopters around until they have a clear "shot"), positive confirmation that traffic conflicts have been acknowledged by pilots, etc. etc.
Larry in TN
En-Route
They get the data regardless of what insurance the owner might have.
The roughly 7.25 million Teslas on the road worldwide are a big advantage to them for data gathering. Each car's autopilot computer is constantly calculating a driving solution and when its solutions differs significantly from what the driver does, the situation is reported back and used in training the AI that produces the autopilot software. They can also poll the fleet for example of a situation that they are working to improve and the fleet responds with examples that meet the criteria.
AV8R_87
Cleared for Takeoff
If they want me to be a test driver, they'd better pay me test driver money. Or at least give me FSD for free.
This car data mining BS has gone too far, and Tesla isn't the only bad player.
EDIT: It's one of the reasons why I hold on to my old car, no such "ET phone home" stuff in it. But when that needs to be replaced, the first mod to the new car will be disconnecting the modem's antenna.
This car data mining BS has gone too far, and Tesla isn't the only bad player.
EDIT: It's one of the reasons why I hold on to my old car, no such "ET phone home" stuff in it. But when that needs to be replaced, the first mod to the new car will be disconnecting the modem's antenna.
Last edited:
AI could be the answer in the future. The ability to measure things so many times per second is so far beyond the human capacity it is staggering. Trying to hit the gap between two arrivals with a departure? AI can measure rates of acceleration/deceleration like no human can, but can they control when the first aircraft turns off the runway so that the departure can be cleared? Can AI advance the throttles at THAT second so separation is assured with the succeeding arrival? It could if you had the capability of AI controlling some/all of what the aircraft does.
That will be a big obstacle as I see it. Before I retired a few years back, there were B757/767 aircraft that still got their 28/56 day FMS updates via 3.5" floppy disks. Some of the MD88s couldn't fly some routes because there were too many characters to load in their nav systems. CPDLC works on some aircraft, but others will never be retrofitted with the equipment. Airlines (rightfully so) don't want to get rid of aircraft they still find useful.
In short, there is such a wide range of equipment capabilities, it makes revolution in the aviation system just about impossible. Evolution is another matter, so whatever is designed, the real trick for the planners is how to evolve it into the NAS.
That will be a big obstacle as I see it. Before I retired a few years back, there were B757/767 aircraft that still got their 28/56 day FMS updates via 3.5" floppy disks. Some of the MD88s couldn't fly some routes because there were too many characters to load in their nav systems. CPDLC works on some aircraft, but others will never be retrofitted with the equipment. Airlines (rightfully so) don't want to get rid of aircraft they still find useful.
In short, there is such a wide range of equipment capabilities, it makes revolution in the aviation system just about impossible. Evolution is another matter, so whatever is designed, the real trick for the planners is how to evolve it into the NAS.
Warmi
Line Up and Wait
Sensor my be 1hz but the logic has potentially full history of sensor reading and can extrapolate future values based on the past and knowledge of what it is tracking and then predict various scenarios.
This is how networking simulation/games are working by predicting future movement of all entities before the actual data arrives on the network to make everything smooth and buffer any minor network issues - it works very well given the the principle of conservation of momentum but obviously can only be used for short intervals - if you loose your network data for too long , the prediction engine will eventually diverge and once you get the data stream back , you may experience your simulation snapping instantly to match the underlying data ( the reality )
I guess what I am trying to say is that there are solutions to most problems but as with everything , result may vary
flyingcheesehead
Taxi to Parking
It would be possible to grow the targets into a zone of uncertainty until the data updates.Sensor my be 1hz but the logic has potentially full history of sensor reading and can extrapolate future values based on the past and knowledge of what it is tracking and then predict various scenarios.
This is how networking simulation/games are working by predicting future movement of all entities before the actual data arrives on the network to make everything smooth and buffer any minor network issues - it works very well given the the principle of conservation of momentum but obviously can only be used for short intervals - if you loose your network data for too long , the prediction engine will eventually diverge and once you get the data stream back , you may experience your simulation snapping instantly to match the underlying data ( the reality )
I guess what I am trying to say is that there are solutions to most problems but as with everything , result may vary
flyingcheesehead
Taxi to Parking
When? Was it the earlier relative/absolute max speed version, or the new improved "either 20 over or 20 under" version?Tesla FSD did get me a speeding ticket.
mandm
Avionics Upgrade
flyingcheesehead
Taxi to Parking
Yeah, the new one (this year) has some issues when it comes to speed. There's a spot I know of where there *used to* be construction. Speed limit is 55, but it still thinks the limit there is 25 despite no signs to that effect. There are other spots where it goes WAY too fast.
But I guess that's why it's supposed to be "Supervised". Elon is a bit more optimistic than I am about when that will happen.
I watched the Daytona 500 on Sunday.
When Elon is willing to strap himself into a stock car with no manual controls and let it run Daytona or Talladega autonomously, then I’ll begin to take self-driving cars seriously.
Better yet, an F1 car at Monaco.
Larry in TN
En-Route
I'm on v12.6.3 in an AI3 car. This version has improved speed control significantly. It will still default to a MAX speed that's too high but you can spin it down with the right scroll wheel and it does a good job of staying to that. v11 was often driving too slow. That seems to be fixed.
Still needs some work on identifying school and construction zones. Not a big deal now, as it's still "Supervised". Will have to be fixed before it can be unsupervised.
455 Bravo Uniform
Final Approach
Does the car have to be competitive or win? Or just finish the race, maybe without a scratch?
Finish at least in top 10 and without damage.
Larry in TN
En-Route
If you ever come through Nashville, look me up. We'll go for a drive and see if I can get you to relax that standard a bit.
FWIW, Tesla provides car insurance in a number of states. Your premium is based on the data from your car so good driving gets you lower premiums. They also offer a discount when the data indicates that FSD is used for a significant portion of the driving. They have data showing a lower rate of accidents with FSD on than with it off.
Lindberg
Final Approach
Why would a race car with no manual controls need a driver? And what do Daytona and Talladega have to do with being able to drive in city traffic?
If you visit central Florida, I'll take you for a drive on I-4 and we'll see if you raise your standards.
This Florida highway is the deadliest in the country. What to know and how to drive safely
Orlando is home to fun theme parks, resorts, and now the deadliest highway in America. An intersecting highway, U.S. 192, is also on the deadly list.
www.news-press.com
Willingness to be a helpless passenger in that circumstance would demonstrate to me Elon's faith in the design.
Well, they should be much safer, no? Cars all going the same direction, no intersections, professional drivers,.... Should be no crashes at all, right?
wanttaja
En-Route
I'd love to see the algorithm that implements, "Rubbin' is Racin'"......
Ron "Rock-em, Sock-em" Wanttaja
Larry in TN
En-Route
I do go to TPA and MCO with some regularity. Not exactly sure what central Florida includes. I'm assuming RSW, PBI, FLL, and MIA are all south Florida. I go there, as well.
FSD does an excellent job on the highway. I can't remember the last time I had an intervention on the highway.
Bacho
Pre-takeoff checklist
Do you think that’s more difficult than getting the car to drive autonomously on public roads? The racetrack sounds very easy by comparison to me.
Lindberg
Final Approach
You could have just said you've never watched a race.
Lindberg
Final Approach
There are around 6 million auto accidents and 40k fatalities in the US every year. Anyone who demands perfection from self-driving cars is just silly.
Last edited:
2-Bit Speed
Pattern Altitude
- Joined
- Nov 9, 2021
- Messages
- 1,604
- Display Name
Display name:
2-Bit Speed
Not sure if he's ever watched a race, but he's been in several...
The last one I can think of in the US was Aeromexico Flight 498 in 1986. Before that, only PSA Flight 182 comes to mind.
Larry in TN
En-Route
Those were mid-airs but I don't think they were ATC error or failure.
PSA was a failure of visual separation. The PSA flight accepted visual separation then ran into the C172. Aero Mexico was a GA airplane that climbed into the TCA without a clearance. The NTSB focused on the limitations of the ATC system's automation and changes were made because of it including the Mode-C vail around TCAs and the push for TCAS.