[Goose]

If you didn't follow the link in Goose's message, you might have missed what I thought was the point of the article.

Ukraine's special forces have developed new software that allows drones to fly without the use of GPS, limiting the impact of Russian jamming.

The software, called Eagle Eyes, allows unmanned drones to travel using sight rather than satellite-based GPS navigation, The Economist reported.

It uses AI to compare live video of the area below the drone to a map made from photos and video that a reconnaissance aircraft previously collected, the report said.
...
The software can also recognize targets, like missile launchers and tanks, and can drop bombs or fly into them without the drone operator needing to give the order,

The Eagle Eye system is similar in concept to the terrain contour matching used by the Tomahawk missile before GPS was deployed. The missile has an onboard radar altimeter that would illuminate the ground and allow the missile to locate itself by the contours of the ground. The ground contours were measured by satellites and could be kept quite current. The Eagle Eye uses natural illumination to allow the software to match views collected by previous overflights. This is order of magnitude more difficult, so doing it is quite an achievement. This approach allows the drone to either fly to a particular position or to follow a programmed course (such as a line of bearing) through the terrain.

If indeed the video is gathered by reconnaissance aircraft or other previous drone flights, it must be problematic. The aircraft and drones gathering the data are presumably GPS denied in the same manner the attacking drone would be. One wonders if perhaps that isn't where the data is coming from :-). It is of course possible the drones/aircraft gathering the data have much better GPS capability than the average "cheap" drone does. I am guessing that the "mapping" software uses AI (or human "I", the way it used to be done) to identify ground features that are easily recognizable and don't move much, like roads, power lines, rail roads, hills, rivers, fences, buildings etc. These features are "extracted" and the pattern recognition AI software in the drone is "trained" to use these reference features to locate itself. I would speculate the software does not work well in a heavily forested flat featureless area. Then again, neither do humans :-).

Using vision to hit a particular target based upon its visual features seems to me to be a different capability than following a map & imagery to a location. 

Using Eagle Eyes, the drone should be better able to put ordnance on or surveil a particular location, without depending on GPS.  (I also expect the maps to have both daylight visual and nighttime infra-red).  If the same or different technology also includes the capability to find the current position of a mobile target, that would be a bonus.  (I definitely agree that decoys and searches make it problematic.  If used unwisely, the enemy can light up a launcher or jammer, shut down & move, and be ready to detect & destroy any drone that comes looking at the old position.)

It is different software technology but it uses the same hardware sensors. Placing decoys near the "real" jammer, close enough to be confusing but far enough from each other that fratricide is not a problem will greatly reduce the probability of hitting the real one. Hunting a mobile target is a "feature" that comes with the ability to recognize the target via software. However searching for it with no real idea which way it went or its intended destination is "trusting to luck".

I think Ukraine wanted to develop a technology that would work with any kind of target, not just one that transmits. The S-400 type of target can be silent most of the time, so relying on its radar to give its location away is not an admissible solution. The S-400 TEL is a very distinctive target and therefore should be easy to recognize. Blowing up the radar is a much better solution though :-).

[Goose]

If you didn't follow the link in Goose's message, you might have missed what I thought was the point of the article.

Ukraine's special forces have developed new software that allows drones to fly without the use of GPS, limiting the impact of Russian jamming.

The software, called Eagle Eyes, allows unmanned drones to travel using sight rather than satellite-based GPS navigation, The Economist reported.

It uses AI to compare live video of the area below the drone to a map made from photos and video that a reconnaissance aircraft previously collected, the report said.
...
The software can also recognize targets, like missile launchers and tanks, and can drop bombs or fly into them without the drone operator needing to give the order,

The Eagle Eye system is similar in concept to the terrain contour matching used by the Tomahawk missile before GPS was deployed. The missile has an onboard radar altimeter that would illuminate the ground and allow the missile to locate itself by the contours of the ground. The ground contours were measured by satellites and could be kept quite current. The Eagle Eye uses natural illumination to allow the software to match views collected by previous overflights. This is order of magnitude more difficult, so doing it is quite an achievement. This approach allows the drone to either fly to a particular position or to follow a programmed course (such as a line of bearing) through the terrain.

If indeed the video is gathered by reconnaissance aircraft or other previous drone flights, it must be problematic. The aircraft and drones gathering the data are presumably GPS denied in the same manner the attacking drone would be. One wonders if perhaps that isn't where the data is coming from :-). It is of course possible the drones/aircraft gathering the data have much better GPS capability than the average "cheap" drone does. I am guessing that the "mapping" software uses AI (or human "I", the way it used to be done) to identify ground features that are easily recognizable and don't move much, like roads, power lines, rail roads, hills, rivers, fences, buildings etc. These features are "extracted" and the pattern recognition AI software in the drone is "trained" to use these reference features to locate itself. I would speculate the software does not work well in a heavily forested flat featureless area. Then again, neither do humans :-).

Using vision to hit a particular target based upon its visual features seems to me to be a different capability than following a map & imagery to a location. 

Using Eagle Eyes, the drone should be better able to put ordnance on or surveil a particular location, without depending on GPS.  (I also expect the maps to have both daylight visual and nighttime infra-red).  If the same or different technology also includes the capability to find the current position of a mobile target, that would be a bonus.  (I definitely agree that decoys and searches make it problematic.  If used unwisely, the enemy can light up a launcher or jammer, shut down & move, and be ready to detect & destroy any drone that comes looking at the old position.)

It is different software technology but it uses the same hardware sensors. Placing decoys near the "real" jammer, close enough to be confusing but far enough from each other that fratricide is not a problem will greatly reduce the probability of hitting the real one. Hunting a mobile target is a "feature" that comes with the ability to recognize the target via software. However searching for it with no real idea which way it went or its intended destination is "trusting to luck".

I think Ukraine wanted to develop a technology that would work with any kind of target, not just one that transmits. The S-400 type of target can be silent most of the time, so relying on its radar to give its location away is not an admissible solution. The S-400 TEL is a very distinctive target and therefore should be easy to recognize. Blowing up the radar is a much better solution though :-).