they were using ardupilot, so the control they gave is "move to this point then descend", latency does not matter much as long as it's reasonable.

In most of the videos I've seen there are failsafe warnings on the screen indicating a loss of GPS, which I'm not surprised at all about. Russia's well-known for having GPS jammers, and having them on-site at an airforce base when the enemy they've been fighting is using drones is just common sense. The video I linked to really looks to me like it's being stick flown with IMU stabilization but probably without Pos Hold.

Why land slowly if the plan is to blow up the drone?

Because you want it to explode at the right location, not get blown off course by a gust of wind or bounce off the wing and explode in the air.

Exploding on impact seems like a very mature, well-established technology.

Exploding on impact is a mature tech for things like shells, but it requires building a mechanism into the shell so that it won't explode before it is fired.

If the drone will be controlled by a human operator till the end, then it might win for the drone design to avoid the complexity of a sensor to detect impacts and of the aforementioned mechanism.

Also, landing on an airplane wing is easier to train for and to test than a mission plan that involves a drone that explodes on impact.

> Also, landing on an airplane wing is easier to train for and to test than a mission plan that involves a drone that explodes on impact.

...and more importantly it is also easier to automate, i.e. autoland on wing spar, detonate, mission complete.