Decentralised and Adaptive Collaboration in Multi-agent Systems

Developing multi-agent systems with self-interested agents with a large behavioural repertoire is a great challenge. In this paper, we pursue the question of how to combine collaborative and non-collaborative agent behaviours in a way to outperform each behaviour individually. In this context, we investigate different collaboration patterns between the agents in a test scenario from the predator-prey domain. Here, the predators move around a 2-D grid and use a parameterisable behaviour model to determine their positions on the grid. Having this behaviour model, each predator can modify a single parameter to change/adapt its behaviour against other predators in the system. This adaptation can be done either (1) in a collaborative way, where the predators collectively search for the optimal set of parameters on behalf of the whole group or (2) in a non-collaborative way, where each predator changes its behaviour individually only on its own behalf. We investigate a method for adaptively combining these behaviours and show how to quantify the resulting behaviour.