Differentiation of Behaviors in Learning Pheromone-Based Communication

Collective intelligence may rely on individuals in a swarm to exploit pheromones as a way to indirectly communicate with each other via the environment. Pheromone-based communication can be effectively exploited in artificial computational swarms to enable simple autonomous agents to collectively coordinate and achieve goals without centralised supervision or control. In this paper, we show how the exploitation of pheromone-based communication - without having to be hard-coded in agents - can be effectively learnt by agents and lead to efficient collective behaviors. In particular, through experiments, we show that: (i) depending on the collective goal to achieve, a population of agents can learn to exploit the same pheromone-based communication mechanism in different ways; (ii) different sub-populations learning concurrently to achieve their own goal - using the same pheromone signal - can effectively coexist with limited interference.