The headset could include discreet lidar/else captor to track eyes movements. Meanwhile the base could track head movements and mouth expression from the outside. Bam you have every data points to animate a memoji. I guess through an API more advanced avatars could be designed by a game developer. But memoji would be a nice proof of concept.
If all this materialize The sweet spot would be a Thunderbolt plug and a M1/M2 requirements. The 3000$ rumored price tag would refer to a complete setup including a mac mini but the headset alone could sell at a more affordable price range.
(Disclaimer: this is a personal guess I have no access to insiders)