AI Just Fully Controlled A Tactical Jet For The First Time Ever

Oxman

Well-Known Member
Skynet anyone?
1676410218365.png



1676410062067.png


A tactical aircraft has made the first-ever flight being piloted purely by artificial intelligence (AI), according to a release from Lockheed Martin. Flying for a staggering 17 hours, the AI-piloted a VISTA X-62A tactical test aircraft that is used to mimic military jets in a trial that hopes to add new capabilities to the VISTA and further automation in military vehicles.


"VISTA will allow us to parallelize the development and test of cutting-edge artificial intelligence techniques with new uncrewed vehicle designs," said Dr. M. Christopher Cotting, U.S. Air Force Test Pilot School director of research, in a statement.

"This approach, combined with focused testing on new vehicle systems as they are produced, will rapidly mature autonomy for uncrewed platforms and allow us to deliver tactically relevant capability to our warfighter."

The AI pilot combined two of Lockheed Martin's autonomous systems, Model Following Algorithm (MFA) and System for Autonomous Control of the Simulation (SACS), to take control of the aircraft while in flight and perform tests “emphasizing autonomy and AI”, though it is unclear exactly what was performed. Autopilots have been capable of maintaining flight and taking over in dire situations for many years, but AI breaking into tactical warfare is a monumentally tougher task.

To test it out, Lockheed Martin used the VISTA X-62A, which is essentially a modified modern F-16D used as a training aircraft. However, it is designated a national asset and there is only one of them in existence, so use of it is limited – the VISTA is also currently being inspected but flights will resume later this year.

Coupled with the serious tech on board, it will now be used as a test bed for AI during controlled flight and in training scenarios, with the potential of training future pilots.

This is a massive deal because finding experienced pilots to train new recruits is no easy feat. While Top Gun may have you thinking that every person training you can take down next-generation jets in a 1970s aircraft, Air Forces are no longer the size they used to be due to extreme costs, so taking ace pilots out of service and into a trainer role is difficult. Should AI help alleviate this burden, new pilots could be trained faster and be of a higher quality.
 
I would recommend Mike Machat's youtube channel. He is an aviation artist and created that Vista livery. He flew with all the GOATs during the 70s out of Edwards, lots and lots of interesting stories.
 
I got into an argument with the Captain about that. It was my leg, I level off at 1500 and he starts yelling "What are you doing CLIMB F'n climb!" I pointed at the MFD and said "we have to be at 1500 for WENTZ". He grabbed his IPad, poked at it all mad, looked at the departure and closed it. He didn't say a word to me the entire ride home. I saved us from being violated.
 
Years ago, as a brand new Citation SIC (first "jet"), PF and climbing out of TEB on one of my very first trips, having fully briefed the departure (including the intermediate level-off), I was so far behind that the IOE Captain reached up and gently pushed forward on the yoke without saying a word, just watching for my realization and correction. One of the best teaching moments I ever experienced, and no ASAP report needed. :cool:
 
I’m a no harm no foul guy but management has come down the last couple months telling us to brasher every RUUDY bust whether or not it loses sep. they’re cracking down on it hard.
 
But wait, I was told by the experts here that AI airplanes was a crazy impossibility.
To be fair what the military considers an acceptable risk is very different from what civilians consider acceptable. We also don't know exactly what the AI was trained to do or how effective it actually was beyond the basics of flight. It is all the curve-balls that get thrown into the equation that is going to make AI take longer to develop.

Personally I do believe it will take some time to develop an AI that matches two pilots in safety. There are a ton of factors in aviation that cars simply do not need to deal with. For example, flocks of geese, random pilots busting airspace that may or may not have a transponder, parts on a plane breaking that doesn't have an input to the AI. Did iceman do his job correctly? "Holy crap the wind just made a huge shift, should I take off?" Is that cell ahead that could cause FA injury going to be avoided if the radar return doesn't show up? What about false returns in mountainous terrain or over cities? Is that ATC clearance actually reasonable or safe? When should I divert? Will ATC give me direct or a route that adds a good 10-15 minutes of flight time due to arrivals? Is that procedure the company came up with actually safe or was it a typo? (Yes, I had MX say I was good to take-off in active snow not knowing if my anti-ice was working on my right wing because a go no-go book said I could...)

Making a dataset to train an AI for the basics of flight is easy. Getting a dataset that an AI can use for all the curve-balls is a completely different kind of task.

Also, will AI actually be cheaper than two pilots? Added complexity and parts means more MX and likely more weight on the aircraft. More weight obviously means more fuel burn. There will need to be a ton of sensors around the aircraft that currently are not there. Not to mention when sensors do inevitably go bad it could likely ground an aircraft immediately rather than be something a pilot can overcome with an MEL.

I do believe AI will eventually take over but there are still a lot of technological advances that will be required. Yes, Tesla does a great job with their automated driving but it only operates in two dimensions with far less to ultimately be concerned about. Hell, they are going to a "camera only" system for autopilot which definitely wont work in aviation.

These are the published limitations of autopilot for a brand new Model S. I guarantee there will be far more for aircraft systems.

Limitations

Many factors can impact the performance of Autopilot components, causing them to be unable to function as intended. These include (but are not limited to):

  • Poor visibility (due to heavy rain, snow, fog, etc.).
  • Bright light (due to oncoming headlights, direct sunlight, etc.).
  • Damage or obstructions caused by mud, ice, snow, etc.
  • Interference or obstruction by object(s) mounted onto the vehicle (such as a bike rack).
  • Obstruction caused by applying excessive paint or adhesive products (such as wraps, stickers, rubber coating, etc.) onto the vehicle.
  • Traffic signs that do not conform to standard recognizable formats, such as digital or temporary speed signs.
  • Narrow or winding roads.
  • A damaged or misaligned body panel.
  • Use of gray or aftermarket glass.
  • Interference from other equipment that generates ultrasonic waves.
  • Extremely hot or cold temperatures.
 
But wait, I was told by the experts here that AI airplanes was a crazy impossibility.

One of the actual experts over at the Microsoft AI program sent me a white paper that showed AI would be able to operate aircraft as well or better than humans do in about 15 years. This was about two years ago. I never disputed it.

What I have been questioning is our society's ability to build and maintain the infrastructure to support AI flying aircraft. I do not believe our society is capable. My friend at Microsoft agrees with me.
 
Back
Top