Game theory turns drone swarms into chase-evasion predators

Researchers Ruiming Zheng, Mohammad Pirani, and Davide Spinello from the University of Ottawa have formulated multi-agent area coverage control as a two-player zero-sum game between cooperative and adversarial agent groups. Traditional coverage control allocates resources based on a fixed environmental risk density field. Here they generalize it by allowing adversarial agents to dynamically generate the spatial risk field, linking both groups through the coverage metric that serves as the game reward. This induces coupled gradient-descent-ascent controllers.

Analysis of a low-dimensional case reveals a Hopf bifurcation dictated by the ratio of each group's control gains. In the adversarial-dominated regime, the system enters a periodic chase-evasion cycle; in the ordinary-agent regime it converges to a fixed configuration. Numerical simulations validate these insights. Under Nash equilibrium, ordinary agents converge to a generalized centroidal Voronoi tessellation, while adversarial agents settle at their own equilibrium centroids. The work has been submitted to IFAC and is available on arXiv.