Research On Mobile Robot Obstacle Avoidance Algorithm Based On Deep Learning

In recent years,with the continuous innovation of artificial intelligence and robotics,more and more scenarios urgently require mobile robots to have the ability to avoid obstacles.In order for robots to be intelligent,autonomous learning is the key to helping them improve their execution strategies and enhance their adaptability and reliability in the environment.Traditional obstacle avoidance algorithms have high requirements for mathematical models,poor actual effects,and lengthy calculation steps,making it difficult to achieve the requirements of safe and fast completion of mobile robots.It is worth mentioning that the convolutional neural network can automatically learn features from large-scale data sets,abandoning the traditional form of obtaining features by people.Applying the convolutional neural network to a mobile robot through the deployed model allows the robot to realize the transformation from mechanical to autonomous,and at the same time allows the mobile robot system to have end-to-end output capabilities.In view of this,based on the in-depth study of the application of deep learning in the field of control,this paper makes a systematic study on the realization of mobile robot obstacle avoidance based on the method of deep learning.The specific research contents of this article are as follows:Firstly,build a mobile robot hardware and software platform on ROS,which realizes the functions of image display,remote control and so on.Secondly,an end-to-end obstacle avoidance algorithm is designed.Firstly,aiming at the advantages and disadvantages of Alex Net and SENet networks,an improved model(multi-connected convolutional neural network)is proposed,and the forward and back propagation process of the model is mathematically deduced.Then,based on the deep learning framework Tensorflow,the classification ability of the multi-connected convolutional neural network was verified on the Cats vs.Dogs,Cifar-10 and Fer2013 test data sets.The experimental results show that the model has good classification capabilities.Finally,the multi connected convolutional neural network is used to train the collected obstacle avoidance data set,and the saved training model is analyzed through training curves,sample tests and so on,at the same of verifying the generalization ability of the model.The network takes the image observed by the robot as an input and directly outputs the steering instructions predicted by the model,including going straight,turning right and turning left.Finally,the trained deep learning model is transplanted to the built mobile robot platform,and the previously trained model is effectively combined with the ROS robot operating system by building ROS?Tensorflow.In addition,by replaying the data from the bag file and dynamically displaying the results predicted by the model in the form of a histogram in real time,the effectiveness of the model is well verified.Through testing in actual obstacle avoidance scenarios,the model's real obstacle avoidance effect and generalization ability are verified.After the environmental information observed during robot movement is calculated by the model,the model can output prediction instructions well.The success rate of obstacle avoidance is 95% and 80% respectively in the real simple and complex obstacle avoidance scenarios,and 65% and 50% respectively in the generalization ability test scenarios 1 and 2.