the DARPA self driving grand challenge was won in 2005, we still have very few self driving cars in the road
Atlas, the robot you linked, was debuted in 2013, it doesn't seem to have been deployed to anything noticeable
fully automated infantry seems significantly harder to me than Level 5 FSD, which seems quite far off---we can't even reliably get self driving cars to work when a human can remote control them
almost all the AI progress lately has been in learning to imitate human behavior in various mediums (text, art, code, etc.; older stuff like image recognition or sentiment analysis also fits in this). in areas where you can't do that progress seems much less imprssive (playing Atari is cool but seems much more primitive than what LLMs can do)
drones are mostly capable of replacing fighter jets today afaik, yet not only do fighter jets still exist, new ones are still being built
the B-52H bomber has been in use by the US military since 1955, the last one was built in 1962, and they are expected to keep flying until 2050. the US military is not eager to phase out components that currently work
afaik, humanoid robots currently cost >$1M to produce. it's quite plausible the typical light infantry soldier isn't worth >$1M. (Musk claims Optimus can get down to $20k, but he has a habit of overpromising so it remains to be seen imo)
in counter insugency operations, a significant part of what armies do is more akin to police work than traditional combat. humanoid robots seem like they'd be very bad at this kind of thing compared to humans