RVO and root motion are sadly two techniques which work very badly together. The reason is that root motion wants to have full control over how the character moves (and only allows it to move in specific and usually relatively limited ways), while RVO also wants full control over the character’s velocity and requires it to be able to change its velocity instantaneously. This requires a compromise somewhere.
Sadly, I have not found a, in my opinion, satisfactory solution for this. One option is to decouple the agent’s position from the transform’s position (ai.updatePosition = false, ai.updateRotation = false) and try to make the root motion follow what the ai is doing as closesly as it can, but that’s not trivial to get to work well.