Research On Swarm Robot Fault Detection Algorithms Based On Mobile Adaptive Networks

Inspired by the collective behaviors of the swarms of insects,fish and other creatures,swarm robotic systems can complete complex tasks using limited communication information between individuals and similar simple distributed control strategies.For the lack of central nodes that can control the distributed nodes,swarm robotic systems are vulnerable to node faults,and non-cooperative behaviors can spread to the whole network through communication,so fault detection is particularly important.Target pursuit is a basic problem in the research of swarm robotic systems,which solves the problem that robots move to a target cooperatively while maintaining the performance of swarm behaviors.Mobile adaptive networks are effective solutions to the problem of swarm robot target pursuit.Based on mobile adaptive networks and the four basic aggregation rules,this paper studies swarm robot fault detection algorithms in the problem of target pursuit.The main contributions of this paper are as follows:(1)The classical fault detection method based on theoretical estimation is applied to the swarm robot fault detection,and verification and analysis are carried out by simulation and realworld experiments in the target pursuit problem.Compared with the original mobile adaptive networks,the swarm robot fault detection algorithm based on theoretical estimation has better performance in accuracy and fault-tolerance.(2)The machine learning methods are applied to swarm robot fault detection.The swarm robot fault detection algorithm based on K-means clustering and the swarm robot fault detection algorithm based on Multi-Layer Perceptron(MLP)classifier are studied.In the target pursuit problem,the training datasets are constructed by sampling in the software defined environment for simulation and real-world experiments.The swarm robot fault detection algorithm based on K-means clustering has a wide range of application,fast speed and strong adaptability to the environment.The swarm robot fault detection algorithm based on MLP classifier overcomes the limitation of the number of faulty robots in the swarm robot fault detection algorithm based on K-means clustering,shows strong adaptability to the type and number of faults,and has high real-time and accuracy,but it needs early data acquisition.(3)According to the different characteristics of the above algorithms,the applicable rules for the above algorithms in different fault situations can be determined.The swarm robot fault detection algorithm based on theoretical estimation is suitable for situations with a great number of faulty robots and low real-time requirements,but the selection of threshold functions is complex;The swarm robot fault detection algorithm based on K-means clustering is suitable for the situations with few faulty robots and high real-time requirements;The swarm robot fault detection algorithm based on MLP classifier is suitable for general situations with the unknown number and type of faults.