AI slays top F-16 pilot in DARPA dogfight simulation

Re:AI slays top F-16 pilot in DARPA dogfight simulation by TuringNYC on Friday August 21, 2020 @08:26AM (#24234281)

I saw the page and assumed the AI had boundaries on maneuvers that reflected the real-world boundaries on those aircraft.

Otherwise, the AI can easily win -- fly at light speed to behind the opponent, fight, fly at light speed away. The only thing to do with a simulation like this would be to put boundaries on the simulator.

[ Parent | on HN ]

Re:AI slays top F-16 pilot in DARPA dogfight simulation by tiku on Friday August 21, 2020 @08:28AM (#24234299)

The boundaries are on the human body, not on the AI. So it's more about speed and certain turns etc.

[ Parent | on HN ]

Re:AI slays top F-16 pilot in DARPA dogfight simulation by TuringNYC on Friday August 21, 2020 @08:46AM (#24234413)

There are definitely also boundaries on any aircraft. There are speed limits, acceleration limits, etc.

The AI is just controlling the aircraft, so any simulation where the AI isnt bound by the aircraft's limitations would not be a good setup.

This isn't different from simulation based training of autonomous vehicles. Imagine if in an AV simulation, the AV controller could opt to accelerate 0-60 in 1 second, or if it could suddenly go 300mph to skip an up coming collision, etc.

[ Parent | on HN ]