Swarm behavior is a fascinating research area. Almost all scientific disciplines feature some interest and publication in this area. The biological origins have been associated a lot with the marvelous behavior and patterns formed by fish moving about in swarms. Some species of birds show similar patterns of collective and somehow coordinated swarm movement. It appears to be tricky to safeguard some form of autonomy in the overriding mass behavior and mass movement.
Recent advances in this field by Li, Li & Zhao (2023) model such behavior within a predator-prey framework to explain the evolution of overall swarm behavior. Whereas “prey” strategies consist in “flocking” and “swirling” to irritate predator(s), the predators make use of “dispersion”, “confusion” as well as “marginal predation tactics”. Additionally, several assumptions go into the modeling of swarm behavioral evolution: agent observation, agent dynamics, and the environment. The mixed cooperative-competitive multi-agent setting applies reinforcement learning, additionally. The framework of evolution has 5 driving forces: (1) homogeneity of a swarm (parameter sharing); (2) actor with decentralized critic network; (3) replay buffer to enhance learning; (4) rewards for predator and loss for prey; (5) learning algorithm.
The modeling approach reveals the impact of an agent-critic network (potential of autonomous behavior) for protection of a swarm in a predator-prey scenario. (Image: fish in fish bowl 2024)































