Research On Efficient Semantic Mapping And Robust Localization Method For Cloud Robotic System

Semantic mapping is crucial for enhancing a robot’s environmental perception ability and promoting the development of intelligent robot applications.However,the complexity of the algorithm and the limited onboard resources of robots have hindered their industrialization and practical application.Cloud robots have emerged as a promising solution to overcome these challenges,allowing robots to leverage cloud resources to enhance their computing and storage capabilities.Nevertheless,the present research on resource scheduling and service guarantee mechanisms for cloud robotics systems is still in its early stages,and it is not yet able to ensure the reliable performance of semantic SLAM to meet the demands of real-world applications.Moreover,the current literature has only superficially addressed the decoupling of the semantic SLAM(Simultaneous Localization And Mapping)system,without conducting a thorough investigation into the decisionmaking theory underlying task offloading between the robot and the cloud platform,as well as data communication within the system.Hence,there is still a need to optimize the costs and task execution efficiency of the system.Finally,leveraging semantic information to improve the perception and behavior capabilities of robots remains an open problem.Therefore,it is essential to explore the advantages and value of semantic information to provide solutions for intelligent robot applications.This paper builds upon the background outlined above to study the resource scheduling strategy of the cloud robot system,the fusion method of the cloud robot architecture and semantic SLAM,and the application of robot intelligence based on the environmental semantic map.The main contributions and innovations of this paper can be summarized as follows:(1)This paper aims to address the challenge of effectively guaranteeing the application performance of cloud robots.In this paper,we initially establish a QoS(Quality of Service)quantitative index system for cloud robot services,taking into full account the computing power and energy consumption of resources at different levels,as well as the influence of communication costs and network delays on the quality of cloud robot services.This paper also conducts a thorough investigation into the correlation between various node collaboration modes and the QoS attributes of the entire task flow,aiming to provide a crucial decision-making foundation for resource scheduling in the cloud robot system.Building upon this work,the paper proposes a QoS-constrained resource scheduling method for cloud robots that enables flexible resource scheduling according to actual application requirements,ensuring the quality of cloud robot services.The experimental results demonstrate that the cloud robot architecture proposed in this paper can ensure that the cloud robot service performance meets actual application requirements.(2)To address the challenge of integrating the cloud robot architecture with the semantic SLAM system,the paper proposes a real-time instance segmentation method based on the cloud robot architecture,which uses the attention mechanism to down-sample the feature map to reduce data communication delays caused by computing offloading,and jointly optimizes the computing offloading point and data compression ratio to enable mobile robots to obtain environmental semantic information in real-time.Building upon this work,the paper proposes a real-time semantic SLAM method for cloud robots,which determines the calculation offload point by minimizing system delay and energy consumption,and utilizes the spatial correlation between geometric and semantic features to reduce the transmission frequency of semantic features,finally achieving the construction of real-time environmental semantic maps on mobile robots.The experimental results demonstrate that the semantic SLAM method proposed in this paper has a low system delay,realizing real-time semantic mapping on mobile robots,and can significantly reduce local energy consumption.(3)Building upon the above constructed semantic map,this paper proposes a robust re-localization method for cloud robots that leverages visual semantic information to effectively handle complex scene changes.Specifically,we propose a scene representation model based on object association that incorporates objectlevel semantic information and inter-object association,resulting in a more robust representation of the scene.We then introduce a scene matching method based on graph matching technology,which uses the random walk algorithm to establish feature matching relationships between scenes,ultimately achieving robust robot positioning in complex scenes.The experimental results demonstrate that the relocalization method proposed in this paper exhibits stronger robustness in long-term operation environments for robots,particularly in scenes where the objects change,when compared to existing methods.By presenting a comprehensive set of solutions for semantic SLAM tasks on resource-constrained mobile robots across architecture,method,and application levels,this paper holds important theoretical significance and practical value in advancing the application of intelligent robots and promoting their industrialization.