You can’t! If by “aim” you mean change the path along which the joint is moving, there is no way to do that. The elbow only does downstroke escape. You could change the motion to some other kind of motion, and in some of your earlier clips up the thread it does look like you’re doing a mix of other joint motions. But in these faster clips it looks more elbow to me. We see the shallow downstroke escape path (diagonal motion) and we see the whole arm moving back and forth. For me, that’s enough for a clinical diagnosis of “successful elbow picking”!
Like @guitarenthusiast says these look and sound great. I would just work on trying to find ways to make this feel smooth. In general, once you have a core motion starting to happen, you enter that longer phase of just using it with as a wide a variety of phrases such that things gradually become more accurate over time, and also more relaxed.
One thing I’d point out is that downstroke escape means only downstroke string changes. So whenever you try to change strings via upstroke, you’ll hit the intervening string. The classic example of this is the “Gilbert” style patterns you’re playing here. Each time you come back from that one note on the higher string, you can hear the little blast of noise as the pick hits lower string. That’s what we call swiping. It’s common. You’ll hear this even on Intense Rock, but it went generally unnoticed for years (decades?) because Paul’s playing in general sounds great.
So don’t worry about this too much. If your playing is generally smooth, fluid, and synchronized, it’s going to sound good. I mention this only because it can help to understand how your technique works. For example, hitting an open string via swipe will likely make more noise than a string that is being fretted or partially muted as it often is in the GIlbert type examples. And also, second, if you play phrases that only have downstroke string changes, those are technically going to be your smoothest feeling and sounding lines, because you won’t be hitting other strings.
I wouldn’t shy away from doing stuff that includes of upstroke string changes if you get good sonic results. I’m just trying to give you clarity so you can more accurately predict / evaluate the results you’re getting.
Nice work here.