| Hexapod robot is a kind of parallel robot,which has multi-degree of freedom,multi-limbs and discrete landing points.With flexible movement mode,strong stability and good environmental adaptability,hexapod robot has broad application prospects.In this paper,a new type of hexapod robot is designed,and the methods of terrain recognition and motion planning of hexapod robot are studied.The mechanical structure,hardware and software control system,forward and backward kinematics,gait planning of hexapod robot are mainly studied.An efficient terrain image feature method and terrain recognition method are proposed.The hexapod robot can recognize the terrain by capturing the topographic environment pictures,and then adopt corresponding motion control strategies to complete the specified motion control tasks.Firstly,this paper introduces the research background and significance of the subject,and introduces the research status of related fields at home and abroad.At the same time,based on the bionics principle,the mechanical structure of hexapod robot is designed.On the mechanical body,the remote host control system,the robot ontology control system and the robot hardware system are built.The software system is designed based on the ROS robot operating system and the idea of layered modularization.Then,the motion planning method of hexapod robot is analyzed and designed.The forward kinematics model of hexapod robot is established based on D-H representation,and the inverse kinematics model of the robot is solved by geometric method.Based on the basic theory of gait design of multi-legged robot,the walking gait of hexapod robot is designed,including triangle gait,quadruped gait,wave gait and rotation gait.Then,the methods of environmental image feature extraction and terrain recognition for the hexapod robot are studied.The terrain image data set Terrain6,which is needed by this research project,is established through data acquisition and data augmentation technology.Based on the transfer learning technology and using MobileNet convolution neural network,a feature extraction method of terrain image is designed to realize the feature extraction of hexapod robot terrain image.Using the image feature set,three classification models,namely support vector machine,naive bayes and random forest,are trained to realize the robot terrain recognition.Based on stacking fusion method,three trained terrain recognition models are fused to obtain the final high-precision terrain recognition model.According to the result of terrain recognition,hexapod robot adopts corresponding control strategy to move.Finally,the simulation analysis and prototype experiment are carried out.Based on the ROS robot operating system,the hexapod robot system designed in this paper is simulated,analyzed and verified,and tested on the prototype of the hexapod robot. |
