Research On The Related Technologies Of Network Optimization And Computation Offloading For Cloud Robotics System

With the continuous development of artificial intelligence technology,the contradiction between humans expectation of robots' function and performance and the limited on-board computational capability becomes more and more intense.Cloud robotics,a novel way to provide small robots almost unlimited resources with respect to data and process,quickly became the most important solution to this problem.However,most of the existing cloud robotics systems focus on the research of cloud intelligence,how to optimize the network architecture from the bottom to ensure the cooperation,tasks allocation and real-time interaction of the robots remains to be researched.Moreover,how to balance the computation processed by the robots and the cloud to improve the task processing efficiency and guarantee the endurance capacity of robots at the mean time needs deep analysis.In addition,most of the existing robot products are costly,which seriously hinders the process of the commercialization and civilian of the robot industry.Therefore,it is urgent to integrate the advantages of cloud robotics systems and provide a solution to deploy AI related applications on low-cost robots.Based on the above backgrounds and the existing foundation of our team,this paper studies the network architecture,computation offloading mechanism and feasibility application of the cloud robotic system.The main work and innovation achievements of this paper can be summarized as follows:(1)Aiming at the optimization of the network architecture of the cloud robotics system,the research is carried out from two aspects:network architecture and network resource allocation.First of all,this paper proposes a distributed construction method of connected dominating set(CDS).Robots communicate with the cloud indirectly through the backbone nodes of CDS,which improves the overall communication efficiency of the whole system and guarantees the robustness and reliability of the system in complex environment.Then,in order to reduce the possibility of network congestion in the cloud robotics system,this paper proposes a network resource pre allocation model and method,which enables the cloud service provider to adopt a more appropriate bandwidth pre allocation strategy,so as to minimize the network congestion and provide network resources to numerous robots fairly.(2)To improve the efficiency of computation offloading method in the cloud robotics system,this paper proposes an energy sensitive computation offloading strategy,which allows robots to evaluate each offloading strategy and select the most appropriate one.In the above strategy,tasks can be offloaded not only to the cloud,but also to other robots to improve the computational capability and execution efficiency.This paper mended the traditional genetic algorithm to find a suitable offloading decision in the shortest time.The experimental results show that the above model and method can let the robot accurately and rapidly judge the feasibility of offloading for each computation task,improve the overall efficiency of the system and prolong the lifetime of the whole network.(3)For the research on the feasible application of the cloud robotics system,this paper realizes the SLAM and the computer vision technologies on the low cost and low performance robots with the help of the computation offloading mechanism.First of all,this paper proposes a multi robot Vision SLAM system based on the strategy of computation offloading,which can offload appropriate computational tasks to the cloud to improve the efficiency of map building and reduce the energy consumption.Secondly,this paper proposes a deep convolution neural network deployment strategy for cloud robotics system.By deploying the deep convolution neural network phasedly,we can achieve efficient,accurate and safe image recognition and analysis function on low cost robot.The simulation results and the actual experiments show that the above two strategies can improve the overall efficiency of the cloud robotics system,reduce the energy consumption of the robots and ensure the safety of network communication.Through the above contributions,this paper can provide a complete set of solutions for the cloud robotics system from three aspects:network architecture,theoretical basis and intelligent application,so as to promote the development and industrial application of cloud robotics and artificial intelligence related technologies.