Research And Implementation Of Robot Intelligent Control Method Based On Machine Learning

In recent years,as the robot application in industrial production and daily life more and more widely,the robot intelligent control method is more worthy of attention,especially the dof humanoid robot walking intelligent control is one of the key research direction in the research field of the robot,the robot movement in the process of inertia and environmental factors,such as a considerable impact on balance of the robot,such as the traditional sports physical model can solve the problem of part of the control,but it is hard to adapt to the requirement of environmental complexity and interactive demand increasing.It is the development trend of robot balance algorithm that machine learning can make robot interact with the environment and constantly improve the balance control ability.Reinforcement learning is a branch of machine learning.As an unsupervised algorithm based on reward and punishment mechanism,it can select the optimal action through interaction with the environment without any prior knowledge.It has strong theoretical and practical value for the intelligent control of biped robots.Firstly,the physical structure model of biped robot walking is analyzed,the reinforcement learning algorithms are studied,and the characteristics are analyzed.Abstract robot agent model,build the reinforcement learning environment.The biped robot is regarded as an agent,and the environment space,state space,movement space and reward and punishment mechanism of the robot are established.Based on the characteristics of continuous movement space output by the legs of the biped robot,the DDPg algorithm is adopted to solve the problem of continuous movement space,and an improved DDPg reward function algorithm is proposed to shorten the training and convergence time,proving the effectiveness and practicability of the improved method.Experimental method:by setting up the simulation experiment environment,the environment of Pyglet and gym simulation library based on Python was built,which restored the real environment information to the greatest extent.Algorithm training was more efficient,and the basic simulation control model was obtained through training.In the real environment,the operation mode of reinforcement learning server and NAO client is proposed.Using Choregraphe and python hybrid programming method,the platform ofreinforcement learning of NAO robot is built to realize real-time transmission of model data,which provides the basis for reinforcement learning of real robot.Experimental results show that nao robot combined with DDPG algorithm can realize stable control in unknown environment.Compared with the traditional control method,this process requires no complicated dynamic modeling process.Finally,after several rounds of training,the biped robot learns to adjust its walking posture to ensure its walking stability.