Research On Mobile Robot System And Search Strategy For Multiple Radioactive Sources Under Nuclear Environment

Nuclear Protection Mobile Manipulator System(NPMMS)has become one of the most widely demanded technologies in the nuclear industry,with the advantages of operational flexibility and rapid movement.The NPMMS can not only be utilized in abnormal inspection and nuclear waste disposal,but also plays an important role in radioactive sources lost/stolen and nuclear leakage accident.Some significant breakthroughs are made in key technologies of NPMMS,but there are still challenges in perceiving and understanding the coupled environment,as well as synchronous exploration with these information.To overcome above difficulties,this dissertation presents the researches on radiation hardness design of NPMMS,multi-source identification for unknown radiation field,independent decision and behavior planning,which have theoretical significance and practical value for improving the intelligence in nuclear exploration task.In order to address the challenges of cluttered obstacles in nuclear disaster scenario,the joint configuration of high-maneuver platform and 7DOF manipulator is developed and analyzed,and the kinematic model of whole robot system is established,which provides theoretical basis for robot navigation and disposal operation.Aiming at the challenges of the severe damage to functional modules by ionizing radiations,irradiation experiment is performed to investigate tolerance limitation of these components.Coupled with actual space requirements,the modular redeployment and centralized hardness design are completed to achieve a robust radiation-proof scheme.Additionally,an universal and multi-level software framework with driver,perception and decision modules is established,on the basis of ROS communication mechanism.The loosely coupled framework improves the real-time performance and modular scalability.In order to address the challenges that no prior information about source number and radiation states exist before exploring the unknown environment,the multi-layer structure of sequential particle swarms,together with Poisson observati--on model and peak suppressed correction,is proposed to predict the radioactive sources,in an online and nonparametric manner.Then the algorithm effectiveness,dimensional scalability and online prediction performance are validated through the corresponding experiments.Considering the extreme condition caused by sparse measurements,Gaussian Process Regression(GPR)is utilized to approach residual strength,which dynamically corrects the swarm weights in both position and intensity aspects.Through the comparative experiment for multi-source prediction,it is further proved that the dynamic correction method has better performance on position error correction,prediction accuracy and optimization efficiency,and that source number and specific states can be timely obtained with both algorithms.In order to address the challenges of the integration between geometric and radioactive information,evaluation fucntions for both physical fields are analyzed and established,respectively.For the geometric model,the unexplored volumes of candidate nodes are estimated as exploration gain,based on Octo Map and traversal query method.For the radioactive model,the non-monotonic curve is presented to satisfy the surrounded sampling requirement,while the index denoting maximum distance ratio between sources is adopted,to eliminate the influence caused by multi-source superposition.Additionally,in order to realize multi-objective explora--tion and behavior planning,the nodes of feasible exploring tree are collected as the candidated states.Therefore,the optimal trajectory and target position can be selected to drive the robot vehicle,with the help of evaluation functions and other navigation costs.In order to verify the system characteristics of both independent exploration planning and radioactive source disposal in complex nuclear environment,the NPMMS is integrated with multiple types of sensors,i.e.radiation,vision and laser sensors.In order to estimate source number and states in the multi-mode radiation field,sampling collections under spiral track and lawn track are utilized on proposed algorithms,i.e.Peak Suppressed Particle Filter(PSPF)and Gaussian residual correction method.The results of experiments show that both algorithms are capable of timely identifying source parameters.In order to verify the effectiveness of the multi-objective exploration strategy,an independent source seeking experiment is carried out for three typical scenarios,i.e.large-scale field,cluttered obstacle field and corridor obstacle field.The experimental results validate the high performance of the proposed strategy,in aspects of multi-task balancing,path planning efficiency and scenario adaptability,and further prove the online performance of the multi--source identification algorithms.In addition,the teleoperated disposal experiments are performed for two typical categories of radioactive sources.The results of experiments show that the developed NPMMS has flexible and responsive operability,good stability and reliability,and broad application prospect.