Moving Decision-Making Of Autonomous Robots Based On The Deep Learning

Recently,autonomous robots have been rapidly developing with its increasing applications,which can be used in many fields aiming to replace heavy,dangerous and poisonous tasks such as polar exploration,rescue exploration,and military aerospace.Autonomous robot is an agent with many functions of autonomous perception,decision-making and execution,and path planning technology is one of the research focuses of autonomous robots.For path planning technology,there are common methods,such as map-building,artificial potential field,genetic algorithm,ant colony algorithm,neural network method and so on.According to the known information of environment,the path planning methods can be divided into global and local.In local path planning,the methods can be divided into traditional and intelligent according to intelligent of the methods.The traditional path planning method can solve the modeling well,but the common problem has insufficient search function on the planning process,usually combining with other search algorithms.However,intelligent methods are good ways to solve the complex environment in path planning,but encounter some problems.Among them,the neural network does not need complex prior knowledge,and has a strong self-learning ability,which provides a new idea for path planning.With the development of artificial intelligence,deep learning has brought many deep neural networks.Neural network method does not need complicated prior knowledge,and has good self-learning ability at the same time,so it provides a new idea for path planning technology.In this paper,aiming to solve the problem of moving decision-making in path planning,a new method based on deep learning is proposed,which combines environmental perception and moving decision-making into one end-to-end way.The way is to directly control the robot's moving direction from the input environmental road information.The main research contents include the establishment of environment perception model with deep networks,the setup of autonomous robot's decision-making model and the research of autonomous robot's moving decision-making on different environments.Above all,the paper transforms the problem of path planning into a common multi-classification,selects convolutional neural network as classification algorithm,setups a deep network model and gets an autonomous robot's environment perception model.Secondly,on basis of the perception model,the moving decision-making model is established,and then the decision-making method is researched.Next,experiments are carried out on different roads to verify the feasibility of the proposed method in the paper.The detail process is as following:First,three cameras placed 30° apart are used to collect the data,or a cellphone collects the image in a similar way and the data are automatically marked.Secondly,a deep learning model is established to can get only three control commands of moving decision-making:turn right,go straight and turn left.Thirdly,do some experiments on the cement,soil,model path,and then compare the three results.Besides,a real-time test is setup in the model road scene on the platform of mobile robots.Among them,the data is automatically marked by the camera placement,namely,images obtained by the left camera is marked as a command of turn right,the middle camera images are marked as a command of go straight,the right camera images are marked as a command of turn left.In this way,the environment information is processed by the convolutional neural network,the classification result is obtained,and the robot moving direction is controlled according to the classification result.Finally,the experiment is completed on the mobile robot platform,and the result show that the proposed method is feasible and effective.