A study in formation control of swarm robotics

Abstract

This thesis presents an algorithm for generating a formation for a swarm of primitive robots. The presented algorithm provide a solutions for both collective decision making and one-to-one role assignment in a connected network of swarm robots. The algorithms can be further be "trained" for a wide range of target applications because the decisions/roles are abstract and lack specific contents. In order to obtain a desired collective shape among simple robots with a small sensor and communication range, formation control techniques are used. This algorithm show how a robot swarm can autonomously choose a loop shape and form the shape in a distributed manner, without central control. Consensus processes, decision making and role assignment, are performed consecutively with a fixed but arbitrary network. Firstly robots will start dispersed in random positions, the swarm aggregate together arbitrary to form a connected network. Consensus decision making is performed with this network fixed, making a collective decision of which loop the swarm should form. Then with the same network, role assignment is performed, assigning the target positions to each robot. After these two consensus processes are done, the swarm disperses and aggregates again, this time aiming to form a loop with robots at their designated positions. The climbing method is used to permutate the robots on the loop. When the robots gets their target positions, they dynamically adjust the local shape so the loop deforms to the target one.