Using AI for flight planning

[hais]

  On 6/15/2023 at 3:29 AM, ilovecornfields said:

make a program that predicts the terminal conditions, checks to see the predicted vs. actual and then modifies the model to try to improve accuracy. It seems that if you did this ever hour for a number of months to years the local model might get pretty good

Expand  

Challenges are computational power and the accuracy of the model. "Classical" weather predictions used models developed by experts. Number crunching them takes significant resources. 

In theory, a ML approach would discover the model. But that would take a large number of parameters and data than what one would collect for an airport for a year. If the underlying model were generic enough such that it would predict accurately for several airports, it might be possible to shorten the time frame needed to collect such data. 

But perhaps the data exists already - it might be worthwhile checking that. One would need to consider though whether parameters that had a certain influence in the past still have the same influence today - climate change would likely be an issue. 

Assuming the data is there, and a meteorologist can provide guidance on what parameters are of interest, your son should be able to use training infrastructure available at AWS, Google, etc. 

 

[ilovecornfields]

Thanks for all the responses. Apparently his group decided to go with a different project but my son did reluctantly agree to teach me Python 3 in his free time so maybe I’ll get to work on this later.

[whiskytango]

Remembering the days when you spoke to a human being to get a Wx briefing, there was an excellent briefer that I often got shortly after getting my IR.  I don't know if he was a pilot, but one of the most helpful parts of his briefing was that he always would say "your escape route will be ... " without ever being asked.  It was reassuring to a freshly minted Instrument pilot to know that someone with a lot of Wx smarts thought there was a safe Plan B if Plan A went south.  It would be nice if AI-based briefings would incorporate that kind of thinking.

[Pinecone]

Heck, you can't even get accurate weather observations some times.

Coming home today, all the local reporting fields were gusty winds.  7 - 12 knots on the low, 20 - 30+ on the high.  Forecast was for the winds to drop.  Winds were around 300 and home field is 1./19.  So decided to give it a try, being cocked to going around and going to a nearby field with a 30 runway and waiting for the winds to drop.

Winds were less than 10, almost down the runway, no gusts.

[A64Pilot]

  On 6/17/2023 at 12:04 AM, 1980Mooney said:

Also GM via Cadillac has 34 million miles on their competing "Super Cruise".  It will be sold on SUV's this year and "Ultra Cruise" is coming in 2024.  "Ultra Cruise, builds off this with a new computing system, that will fuse the incoming data streams into a unified 360-degree view around the vehicle."

The nay sayers need to just get out of the way - lest they be the ones run over.....

Expand  

Unless things have changed Super Cruise only works on mapped roads, so it’s not close to true self driving.

We have had our Tesla for a couple of years, I don’t have full self driving, but all come with Auto-pilot, which is a very limited version, Full self driving is available, for $15,000 and it doesn’t work, but has been promised this year, for years. I’m not dropping 15K hoping that finally this will be the year, get back with me if or when it works, yeah I know then it will be $30,000, that’s free enterprise.

Most, almost all confuse auto-Pilot that ALL  Tesla’s come with for full self driving which is I’m nearly certain still Beta and not many have

https://insideevs.com/news/666082/tesla-musk-new-plan-full-self-driving/#:~:text=Tesla continues to release incremental,it's not called version 12.

Tesla has built this huge mega expensive computer and as all cars are connected to Tesla it’s supposedly learning from their mistakes, it’s my understanding that the theory is Tesla will be full 100% AI and only use vision, no radar, Lidar etc, only cameras.

Auto pilot is pretty darn close for me, it won’t navigate or stop at signs or lights, but it’s pretty darn good driving, and brilliant on highways, because highways are easy, no crossing traffic, no on coming traffic etc, etc. City streets and especially neighborhoods are actually very complex

Personally I don’t think we are even close to full self driving, not unless we change roads. Right now if your driving in a neighborhood and a baseball rolls out in between two parked cars, you will hit the brake, not because the ball is a threat to the car, but because you know it’s likely a child is chasing it, I don’t think anybody’s self driving computer is even close to putting that together. I don’t think any of them will avoid a pothole, even a wheel busting one for example.

I’m not even close to being a computer guy anymore, but my understanding of true AI is not a program that responds to events in accordance with its program, but one that will respond to events that it’s not been programmed for, and if it gets it wrong then it will learn as in I guess modify its programming so that next time it gets it right, and will even be able to make judgement calls, like if it’s driving a bus, does it run over the kid in the road, or run the bus off a cliff?

I think Musk has his eye on way more than cars, cars I assume is just the start? He’s hiring by the way

https://www.tesla.com/AI

Edited June 19, 2023 by A64Pilot