The evolution of behaviours in swarms of robots

Evolutionary swarm robotics uses evolutionary computational techniques to synthesise behaviours for a group of autonomous robots. In a swarm of robots, the collective behaviour of the robots results from the local interactions between robots and interactions between robots and their environment. The design of such system aims to exhibit the same characteristics as simple biological systems: simplicity, robustness, flexibility, and modularity. This thesis presents research on the evolution of individual behaviours in simulated mobile robots using genetic algorithms that collectively exhibit robustness. To explore these characteristics, the evolved behaviours are not only examined in changing environments, but also in models with a varying degree of abstraction. These models include an abstract model (2D, noise free environment), a realistic model (3D, noisy environment, takes physics laws into account), and a real-world model. The behaviours evolved in simulation are then transferred onto robots in the real-world in order to analyse the reality gap, a common problem in evolutionary swarm robotics. A number of experiments are carried out in order to measure the robustness of the system. The first examines the feasibility of evolving behaviours on simulated robots with limited capabilities and the effect of noise on these behaviours. The second experiment set inspects the reality gap between the abstract (2D) and realistic (3D) simulators as well as the effect of noise in different environments. The last experiment set investigates the reality gap between the realistic simulation and the real-world model.

Funder

Higher Education Authority

Publisher

NUI Galway

Rights

Attribution-NonCommercial-NoDerivs 3.0 Ireland

cbn

Collections

University of Galway Theses (PhD Theses)