Swarm-Robot Distribution Control And Optimization

Swarm-robot system is one of the important research directions of multi-robot systems, whose major feature is that the system includes many robots. How to control swarm formation in unknown environment is a challenging problem. In this paper, Swarm-robot distribution control and optimization is studied. The main work and research results lie in the following.1. We introduce swarm robotics definition, characteristic and the art of development. We also analysis swarm robotics theory and method, summary swarm robotics main research content.2. This paper analyses the principle on grid distribution and motion control of swarm-robot based on virtual force, putting forward some performance judgment parameters (such as stabilization time, collision times, connection number, cluster number, and so on). The effect of virtual force on grid formation motion is discussed. The result of the experiment shows that this method can accomplish the grid distribution and motion control robustly and efficiently. We also study the stability properties of the collective behavior of the swarm.3. On improving the performance in which the swarm-robot grid formation motion are controlled in a complicated circumstance based on virtual force method, it is used the multiobjective genetic algorithm to optimize control parameters. Performance indexes include collision,break the ranks,connectivity, etc. The weight value of the indexes is determined by their importance. Optimization model is established, the multi-objective genetic algorithm based on Pareto sets is used to search the solution of the problem. Simulation results show that this algorithm is effectively capable of obtaining a set of non-dominated solution within a finite evolutionary generation, which overcomes the weakness of handiwork to set control parameters.4. Design a paradigm that each robot can online learn and change information among robot based on multi-agent distribution reinforcement learning. This online learning paradigm has the ability to allow robots to learn and adapt to unexpected scenarios in new environments. Simulation Show that the presented algorithm based on process reward can decrease interference, avoid deadlock and improve group performance. Resolve the shortcoming of the Noise of environments and non-existence of a global observer.5. Inspired by the physics paradigm, we discussed swarm-robots system capturing principle and method that based on virtual force. We provide some performance metrics (such as stabilization time, layer number, distance between layers, robot number in a layer, distributing density and so on); Analysis the relation between the capturing performance and the robots number, the virtual force. Put forward the layer force structure which not only quickens the capturing speed, but also improves the quality of capturing layer. Put forward the capturing method that robot moves around the target, the result of the experiment shows that this method can solves the local minima problem, and greatly improves the capturing speed and quality.6. We have designed a swarm-robot simulation system. This is a digital system that is used for the demonstration of the control architecture, cooperative control, learning algorithm of swarm-robot system. This paper discusses the physical feature of each real part and the implement difficulties, presents the software architecture and the object model, and describes the implementation of each function. Result shows that it has good performance, also it works well in distributed simulation environment and has good interface to interact with the users. It offers an effective platform for the swarm-robot system research. We also design a physics robot experiment to verify swarm-robot distribution.