Text this: Residual Learning for Riemannian Motion Policies in Multi-Drone Obstacle Avoidance and Communication-Aware Navigation.