Marine Armor tried that some years ago at 29 Palms.
The big glitch was terrain. Human eyes can still see depth and terrain better than any video feed. The tanks kept getting stuck or tracks damaged by rugged terrain. An overwatch was used but still missed critical terrain features.
When the way forward is uncertain, controlled driving will be slower. Slow means dead.
A human driver can blend 3D vision with speed judgement.
You can now get stereoscopic video rigs from the commercial telepresence industry that just about replicate said vision, something not an option back then.