Research On Socially Aware Navigation Of Home Service Robot Based On Personalized Comfort Space Model

With the development of artificial intelligence technology and home intelligent space,robots are gradually coming into people’s lives.Socially aware navigation provides a new idea for home robots to realize human-centered navigation.The so-called socially aware navigation refers to the completion of the navigation process based on social cues related to people and the environment,and to improve the comfort of navigation in the process of human-robot interaction.Socially aware navigation requires the robot to maintain a comfort social distance in addition to a safe distance from people.Unlike public indoor scenes such as airports and shopping malls,home scenes have a small number of pedestrians,relatively fixed identities and long-term interactions with robots,and differences in age,gender,and closeness to robots among different family members,which lead to higher individualized requirements for comfort distance.Therefore,this thesis addresses the problem of socially aware navigation for home service robots.First,a personalized comfort distance for different users is inferred based on the fuzzy systems.Then,based on this,a comfort space for users is constructed using a mixed Gaussian models.Finally,the comfort space is constructed as a comfort space layer of a layered costmap,realizing socially aware navigation that meets users’ personalized comfort space requirements in the process of home robot service.In addition,an intelligent space human-robot interaction system is built to help users understand the navigation behavior of robots,further improving the comfort of socially aware navigation.The main work of this thesis is as follows:Aiming at the personalization problem of socially aware navigation to keep the comfort distance,a fuzzy system-based user personalized comfort distance inference method is studied to meet the personalized demand of home users for comfort distance.First,the proxemics theory is combined with the types of home service tasks to roughly classify the comfort distance.Then,for high-frequency and user-sensitive interaction tasks,a fuzzy inference system based on a fuzzy tree is proposed to obtain a more refined comfort distance.The inputs to this system are multiple social cues obtained by the robot from the user and intelligent space,satisfying the need for personalized modeling of user comfort distance.With the accumulation of data on human-robot interaction in the home,the fuzzy tree is optimized based on global and local optimization methods to achieve personalized learning and correction of comfort distances,improving the intelligence of the robot.Aiming at problems of insufficient personalized expression of the traditional personal space model,insufficient fine description of the comfort space and low efficiency of robot navigation,a mixed Gaussian comfort space model construction method based on personalized distance is proposed to optimize the balance between comfort and navigation efficiency.First,the user’s personal space is finely divided into six directions,and modeled as six standard deviations of a mixed Gaussian model.The standard deviations are adjusted dynamically based on physical and psychological factors as the user’s speed changes,resulting in a dynamically changing comfort space.Next,a group comfort space fusion strategy is proposed to ensure that robot navigation does not disturb interactions among group members and conforms to social norms.Finally,navigation target points are generated based on both individual and group comfort spaces,providing comfort service locations for delivery-type services.Additionally,an area-based evaluation index is proposed to compare with traditional personal space,reducing redundant space occupancy and improving the efficiency of human-robot collaborative tasks.Aiming at the lack of comfort of robot navigation in the home environment,a socially aware navigation framework is proposed to realize the application of socially aware navigation in the service process.An intelligent space human-robot interaction system is designed based on PyQt,which renders a 3D model of the home environment to reflect the robot’s actual location and planned path in real-time,and displays information such as the current task being executed,service knowledge and task sequences based on an ontology library,and real-time camera feed from the robot,in order to alleviate users’ fear of unknown robot behaviors and further improve the comfort of socially aware navigation during the service process.