Behaviour-based pattern formation in a swarm of anonymous robots

Abstract

The ability to form patterns is useful to maximize the sensor coverage of a team of
\nrobots. Current pattern formation algorithms for multi-robot systems require the
\nrobots to be able to uniquely identify each other. This increases the sensory and
\ncomputational requirements of the individual robots, and reduces the scalability, ro-
\nbustness, and 
\nexibility of the pattern formation algorithm. The research presented
\nin this thesis focuses on the development of a novel pattern formation algorithm
\ncalled the Dynamic Neighbour Selection (DNS) algorithm. The DNS algorithm does
\nnot require robots to be uniquely identified to each other, thus improving the scal-
\nability, robustness, and 
\nexibility of the technique. The algorithm was developed
\nin simulation, and demonstrated on a team of vision-enabled Bupimo robots. The
\nBupimo robots were developed as part of the research reported in this thesis. They
\nare a low-cost, vision enabled, mobile robotic platform intended for use in swarm
\nrobotics research and education. Experiments conducted using the DNS algorithm
\nwere performed using a computer simulation and in real world trials. The exper-
\niments conducted via simulation compared the performance of the DNS algorithm
\nto an other similar algorithm when forming a number of patterns. The results of
\nthese experiments demonstrate that the DNS algorithm was able to assume the de-
\nsired formation while the robots traversed a shorter distance when compared to the
\nalternative algorithm. The real robot trials had three outcomes. First, they demon-
\nstrated the functionality of the Bupimo robots, secondly they were used to develop
\nan effective robot-robot collision avoidance technique, and lastly they demonstrated
\nthe performance of the DNS algorithm on real robots.