Formation Control of Localised and Decentralised Robotic Swarms

Abstract

Robot swarms consist of multiple autonomous robots, which detect and interact with their local environments. The fundamental intelligence is observed when a chaotic swarm reaches a stable and orderly objective formation. The process is important because the objective formation is designed such that the swarm achieves more than the sum of its individuals. A formation is a set of positions or tasks, and intelligent swarms are capable of self-organising and task allocating. Given an objective formation, individuals of a swarm reach different objective positions and perform different tasks. This implies task allocation in different possible choices. For each individual, the path to its objective position is regarded as the efort to take, and the inclination to different objective tasks means different eforts. The challenge is that it needs to choose wisely in the interaction with its neighbourhood. Changes of choices are compromises and each progress to the objective position imposes in uence on its neighbourhood. The collective intelligence comes from series of individual decisions in the process. In this thesis, we consider four problems that arise with the challenge. We use techniques from graph theory and agent-based design to address them. Formation control algorithms should not impose heavy burden in the communication network. Thus, to start with, limited sensing and communication are assumed, and the robots have minimal access to each other's identity through locally established channels. The control strategy is proposed based on local optimisation and multi-object mapping for a team of robots. Robots are able to make mapping decisions based on local information. To achieve the local optimal mapping decisions for each robot, two novel multi-object mapping protocols are designed. The first protocol performs confict locating and resolving, and the second adopts a most-neighbour mapping strategy. The formation problem is further addressed for a scalable team of robots subject to limited sensing with no communication. The robots themselves are fully independent with no designated roles. Scalable objective formation design is proposed such that the robot formation is scalable. Under the assumption that the data transmission among the robots is not available, a novel controller and a protocol are designed that do not rely on communication. As the controller only drives the robots to a partially desired formation, a distributed coordination protocol is proposed to resolve the imperfections. The case is investigated where the objective formations are arbitrary and have fixed sizes. Multi-objective mapping is proposed for the individual robots to identify their positions in the objective formation. The fixed formation size induces mapping loops, and to avoid local optimum traps, an evaluation method imposes a weak restriction on the predened formation, rendering it almost arbitrary. To enhance the robustness, the minimal local topology is proposed, and to reduce the computation burden and avoid the infnite trajectory loop, the coordination protocol is modifed by introducing probability. The practical problem of collision avoidance is also studied. The leaderfollower scheme is implemented on a multi-robot platform. On the premise of coordinated control laws, globally desired formation is achieved. The same problem in the path-planning perspective is considered on a global scale. Disc obstacles are filtered and clusters are identified based on their intersections. The path planning algorithm is designed based on obstacle clusters.