Research And Implementation Of Biped Teaching Robot Development Platform

With the rapid increase of computing power and the mutual penetration and continuous development of subject technology and the continuous innovation of the teaching mode under the background of “New Engineering”,the development and implementation of biped robot as the carrier combines mechanical control,sensor development,image processing and neural network learning.The multi-disciplinary and cross-integrated teaching experiment platform of the function bridges the gap between the experiment and the practical application of the Internet of Things and artificial intelligence.A biped teaching robot development platform,which combines the experimental learning of automatic control,computer language programming,programmable logic gate array design,embedded system and artificial intelligence was designed.The main research contents of the thesis are as follows:Firstly,the design requirements and the design content of software and hardware coordination are analyzed in depth,and the execution structure of the biped robot and the selection of the main controller are determined.In the system design scheme,considering the development of the language environment,hardware scalability,algorithm implementation and hardware acceleration,the heterogeneous processor platform PYNQ-Z1 was selected as the main control board of the robot.Secondly,a custom hybrid library of biped teaching robot based on PYNQ framework is constructed.The design of the hybrid library is implemented from the theoretical point of view to the FPGA hardware implementation,which can complete the attitude setting of the biped robot,gait planning and image classification based on convolutional neural network.Firstly,the kinematics model of biped robot is established and the forward and inverse kinematics equations are derived.The gait planning based on zero moment point constraint is simulated and the trajectories of hip and ankle joints are simulated.The Arduino soft core system is proposed.Robot control program.The high-level synthesis tool is then used to design and optimize the convolutional neural network accelerator IP core for different scales and structures,and the call interface is encapsulated in the Python language.Finally,an experimental scheme for the development platform of the biped teaching robot was designed.On the Juypter Notebook webpage,the attitude and walking control of the biped teaching robot are completed.The camera is used to collect the handwritten digital character pictures in the forward field of view and input into the convolutional neural network accelerator for forward prediction.The result of image classification and recognition can be characterized as the command meaning as the walking instruction of the biped robot,so that the robot can recognize and interact with the external information.The experimental results show that the mechanical control,neural network learning,image processing,Python language programming and other functions of the biped teaching robot development platform have reached the standards of high versatility and openness,which meets the learning requirements of multidisciplinary cross-integration.