The Research And Implementation Of Software Architecture And Techniques In Human-Computer Interaction Of Humanoid Robot

Humanoid robots are similar in appearance and movement to human beings,can be used without changing the living environment of human beings,and it is also quite acceptable for users.Therefore,it is suitable to use as a kind of service robots.Humanoid robots involve many key technologies,including artificial intelligence,machine learning,human-computer interaction,which will need a complicated software system.At present,the software systems of humanoid robot are suffering from problems of high cost and poor portability.At the same time,the controller based on the reinforcement learning may requires of real-time performance and simulation compatibility to the software system,which puts forward further requirements for the software architecture.There are many restrictions on the operating system and programming when developing real-time task using traditional methods.In addition,when using humanoid robots in everyday life,a good human-computer interaction method,such as such as gestures and natural language can enhance the user experience.In the gesture recognition,the extraction of the hand area is the prerequisite and difficulty of identification.In understanding the natural language,the understanding of the reference in dialogue is necessary.In view of problems on software system of humanoid robots,this paper proposed humanoid robot software architecture base on ROS,which running Ubuntu operation system using Low-Latency Kernel.This architecture could provide the ability of cross-platform and cross-language developing,which will reduce the limitations of the development environment and make it convince to embedding simulation software and other open source codes.Aiming at the limited ROS support environment,this paper proposes a communication method based on bridge nodes,so that the general process without ROS supported can also use ROS topic and service.In addition,some solutions for the problem of human-computer interaction are presented.Firstly,this paper proposes a hand region extraction technique based on skin color model and binocular vision,and then uses the simplified HOG feature extraction algorithm and SVM classification algorithm to classify the gestures.The provided technique could achieve a good effect when the training set is limited and the number of gesture is slightly large.Secondly,this paper uses semantic diagrams and vision diagrams to represent the information about natural language and computer vision,and then provides an optimization method to finish the inexact graph matching to make the robot understand the reference in dialogue of nature language,i.e.reference resolution.Finally,a number of experiments are conducted to validate the software architecture,and demonstrates a human-computer interaction scenario using gesture recognition and reference resolution,which could verify the role of these methods.