Research On Joint Communication-motion Planning In Multi-Robot Systems

With the rapid development of artificial intelligence and automation technology,robots are being widely applied in our daily lives.Compared with single-robot systems,multi-robot systems(MRS)have the advantages of robustness,parallelism and distribution,which can also improve the task performance by cooperation.Multi-robot cooperation involves large amount of data and control information exchange,which puts forward high requirements for wireless communication quality.For the MRS,there are two main kinds of approaches that can improve the communication performance.The first approach is based on the traditional communication planning,and the other improves the communication performance by adopting the robots' mobility which may also incurs extra performance loss,e.g.,motion energy consumption.Therefore,taking advantage of communication and motion resources to make joint communication-motion planning(JCMP)is meaningful for task performance improvement.In this thesis,we focus on solving JCMP problems in different scenarios,and the main works and innovations of this thesis are as follows:· Designing the MRS-oriented JCMP problem solving framework(Chapter 2)This thesis firstly proposes and analyzes the JCMP problem in MRS.Based on the cognition cycle and robot behavior models,we design the JCMP problem solving framework.According to the Boyd's OODA loop,we divide the framework into four stages: environment sensing,performance evaluating,planning and policy executing,and design the corresponding function for each stage.The framework can provide a methodology guidance for solving JCMP problems,and researchers may achieve rapid and efficient system development according to their needs.· Proposing a JCMP problem solving method for the scenario without prior task information(Chapter 3)This thesis considers a multi-robot surveillance scenario in an unknown environment.In this scenario,a team of robots are responsible for surveying an outdoor area and transmit the monitored data to the base station through a relay robot.In order to minimize the MRS's energy consumption as well as guarantee the link quality between the patrol robots and the base station,the relay robot should ad-just its positions according to the movements of the patrol robots.This scenario involves the joint planning for the robots' transmit power and motion trajectories.Firstly,based on the end-to-end packet error rate(PER),we derive the asymptotic closed-form expressions of the optimal transmit power of the relay robot and patrol robots.Then,we model and solve the JCMP problems in this scenario according to the framework proposing in the second chapter.In order to fully utilize the sensing abilities of the patrol robots,we consider a more general scenario where the task performance can be further improved.We test our method in different system settings,and the experiment results show that our method can significantly save the total energy consumption than benchmarks.· Proposing a JCMP problem solving method for the scenario with prior task information(Chapter 4)When having prior task and environment information,we may try to obtain a longterm performance metric by jointly planning the communication and motion resources.We select the channel capacity as our long-term performance metric,and consider two typical conditions: finite-horizon and infinite-horizon problems.In the finite-horizon problem,we focus on a multi-UAV(unmanned aerial vehicle)scenario where we aim to obtain the relay UAVs' optimal trajectories based on dynamic programming(DP).In the infinite-horizon problem,we consider a multiheterogeneous-robot system.In order to overcome the curse of dimensionality of DP,we adopt the approximate dynamic programming techniques to optimize the computation and memory complexities,such as feature extraction,function approximation and state space sampling.We verify the effectiveness of our method by comparing it with DP and the greedy method.· Proposing a JCMP problem solving method based on sensing hardware(Chapter 5)The channel models based on theoretical parameters can hardly describe the link quality for specific scenarios.Therefore,we consider modeling the wireless channel based on measured data.The main measurement-based channel modeling methods can be classified two types: model-based probabilistic channel modeling and interpolation-based radio environment map(REM)modeling.For the first modeling method,we design an autonomous environment sensing hardware platform on ROS(robot operating system)framework based on GNU Radio and USRP(universal software radio peripheral),which can autonomously measure the channel quality in different frequency bands.Then,we model the wireless channel based on measured data,and optimize the MRS's throughput by taking advantage of the resources in frequency,space and time fields.Finally,considering the time-varying characteristics of the spectrum,we propose a fading-based JCMP solving method,which can further improve the system's throughput by adopting the fading effects.On the other hand,we focus on the REM modeling method by adopting the Microhard P900 modules.We consider a delay-reliability tradeoff problem for wirelessconnected indoor robot surveillance based on REM.Based on the deep analysis of the problem,we propose an efficient patrol path planning algorithm with pruning and connectivity judging.The experiment results prove that our method can greatly improve the efficiency of the optimal path search on the basis of satisfying the communication delay and reliability requirements.