Self-driving cars are hard
Self-driving cars are actually hugely difficult to make. Imagine a camera feed. You have a grid of pixels. Essentially a huge chequerboard full of binary digits.
Now try to write an algorithm that can take that chequerboard and turn it into a meaningful map of the surroundings. It's a very hard problem.
Going off road multiplies the problem
If you are on a road, you can reduce the scale of the problem considerably. You are surrounded by rectangular cars. Slow moving upright objects are probably people. Converging light coloured lines are probably roads.
Going in the air makes it easier still. There is almost nothing to hit. All you need to do is proceed to a GPS coordinate.
In a tank none of these assumptions hold true. Tanks are camouflaged. They may be partially occluded. You might be in a forest or a destroyed city. There might be obstacles.
Now add in tactics
The bridge has been damaged and might be dangerous to cross. The trees that should offer cover have no leaves because it's autumn. The enemy is flying a false flag. There are RPGs on that hill over there, covering the road.
A human with common sense (learned heuristics) can make these complex, unpredictable decisions. A machine can't, because machines can only do what they are programmed to do.
What about a learning computer?
You might try to make a computer that can learn from its environment. This is a very hard problem indeed. You now have to take your chequerboard input and somehow convert that grid into "rules for staying alive". No one is even close to being able to write a program to do this. Such a program may not even be possible. We might need quantum computers, or something exotic we haven't yet thought of.
All current machine learning algorithms require carefully curated datasets, and hand tuned algorithms. You can't just chuck in a real world, noisy video image at them.
Our current best computers give us no more than insect level intelligence if that. You would need mammalian level intelligence to power a free roving autonomous tank.
People are clever. Computers are dumb.
So what about remote control? In the air, if your signal is jammed, you can simply fly to a preprogrammed GPS coordinate and land. On the ground, if you lose your signal, you are on your own. You can't drive home in a straight line. Plot a predictable route home and your unpiloted convoy could be taken out tank by tank by a guy making holes with a mechanical digger, or some similarly unpredictable obstacle.
We may one day be able to make a computer intelligent enough to function autonomously in the world without a human guiding it. That day is a long way off and we have no roadmap of how to get there. We really have no idea about what intelligence actually is, or what problems we need to solve to reach it. True AI is still no more than sci-fi.
(Source: I studied AI at Sussex University.)