Sorry if I sound silly saying this, but why build a robot on one end and put a human on the other, when you could put a human on both ends of the avatar project and build the robot around the two?
Point is, put a capable human in place across the world with a suit on that is attached to a human controller with another suit on at the control station. The suit the controller wears would send signals to the suit the flesh-human avatar wears at the remote location. Then it would only be a matter of gradually replacing the biological parts of the avatar with robotics. Think haptic feedback.
This way you could start with practical applications like performing surgery from a distance using multiple human-trained avatars on location to do heart surgery for example. Both ends would have HUD, cameras, and haptic feedback. Audio cues would go through the nervous system of the two organic pieces of the puzzle. Real world applications would be where you start, and not with the robot.
It’s the Tony Stark Ironman method. Build the suit around the body until the suit doesn’t need the body. So you take out the body at that point and do the reverse, filling the suit with the body of the robot avatar. Lastly, remove the suit. Now you have your avatar.