Research On The Active Interaction Technology Of Humanoid Robot Based On Intention Perception

The active interaction mode of humanoid robot can further narrow the distance between human and robot,and improve the naturalness and harmony of human-robot interaction.The perception of human interactive intention is an important basis for active interaction decision-making of humanoid robots.Interactive intention perception allows interaction agents to make quick decisions based on the predicted results so that the robot can better respond to the needs or mental states of users.The quantitative recognition of interactive intention is the key to distinguish the interactive intention of different users.However,most of the existing interactive intention recognition methods are qualitative,that is,only identify whether the user has interactive intention.Therefore,this thesis proposes a quantitative recognition method of interactive intention intensity,which enables humanoid robots to make reasonable active interactive decisions based on the perception results of different objects’ interaction intentions.The main works of this thesis are as follows:(1)A set of humanoid robot active interaction system is designed,which mainly includes three key steps: potential interaction feature extraction,interactive intention intensity inference and robot active interaction decision-making.(2)In terms of potential interaction feature extraction,first,the visual features that have strong correlation with interaction intention are explored and analyzed,and two types of transient features(linear distance,body orientation offset)and one type of temporal feature(motion states)are selected for interactive intention intensity recognition.Second,the 3D skeleton information of the pedestrian is extracted using the multi-person pose estimation method,and two kinds of transient features are further calculated.Finally,a set of pedestrian motion states recognition dataset based on 3D skeleton sequence is made,and an efficient pedestrian motion states recognition network based on HSTAM-GRU is designed to achieve the extraction of temporal feature.(3)In the aspect of interaction intention intensity inference,first,based on the multidimensional social experience,two sets of inference rules for static and dynamic interactive intention intensity are designed,which achieve the fusion of multidimensional interactive features.Furthermore,a bi-state rule fusion method is designed to fuse static and dynamic inference rules.Then,based on Mamdani fuzzy inference algorithm,pedestrian interactive intention intensity inference is realized;Finally,the center of gravity method is used to de-fuzzy the inference results,and the precise intensity value of interactive intention is obtained.The experimental results show that this method has excellent performance and can effectively perceive the interaction intention intensity of different pedestrians.(4)In the aspect of robot active interaction decision-making,first,based on the interactive intention intensity recognition method,the main interaction object selection mechanism and interaction behavior trigger rules of humanoid robot are established.Second,based on serial communication,the active interaction control algorithm is designed and carried on the humanoid robot platform;Then,experiments are carried out to explore the optimal interactive intention intensity threshold when the humanoid robot’s active interaction is triggered.Finally,the effectiveness of the proposed active interaction control algorithm is verified by multi-person active interaction experiments under the condition that the optimal interaction trigger threshold is set on the humanoid robot.