Multiple drones, operating as a swarm, can be applied in many situations, including emergency response and mapping. However, it is difficult to coordinate drones that fly close together because it is hard to capture the interactive aerodynamic forces generated by each drone. In this paper, Shi et al. present Neural-Swarm2, a machine learning method for motion planning and control of dense, heterogenous swarms. Their DNN-based approach uses the relative position and velocity of neighboring drones to predict the interacting forces, which are also used to define a motion plan. Neural-Swarm2 can enable the operation of drones with lesser vertical proximities and significantly reduces tracking error compared to other control methods.