In recent years, the Smoothed Particle Hydrodynamic (SPH) method has been successfully applied to model swarm robotic systems as incompressible/ compressible fluids. Essentially, the SPH approach models inter-robot interactions using attraction-repulsion force profiles and in this respect is reminiscent of traditional analytical frameworks used in swarm systems such as Artificial Potential Field based methods. However, in contrast to other virtual force based approaches, the SPH method provides a much more effective way to control the density of the robots; a particularly useful feature in several applications of swarm systems including pattern generation and coverage control. In this paper, we revisit the SPH method from a control point of view with an emphasis on density control, and propose the idea of density-based control for multiple robots. In addition, we modify the original SPH method by fully decentralizing the SPH controller while retaining its density control feature, and introducing an inter-robot collision avoidance mechanism. This enhances the capability of the model in controlling a swarm of real-world robots. Finally, the effectiveness of our density-based control of a large number of robots is demonstrated through implementing two important tasks in multi-robot control: group motion and shape control, and group segregation.