Research On Active Scene Recognition And Path Planning For Mobile Robot Of Indoor Scenes

With the development of artificial intelligence technology and social demand,mobile robots are widely used in the national defense industry,public security,disaster relief and family services.For service type indoor mobile robot,accurate classification of scene types is the premise for other family work,and quickly adapt to the complex indoor layout and complete the path planning is the key to realize the family service.Therefore,indoor scene recognition and path planning of mobile robot technology have great research significance.Paper carries out experimental research on active scene recognition and path planning.Firstly,for robot indoor experiment,we design the 3D indoor simulation environment platform in Gazebo simulator,and laser radar,RGB-D cameras and other sensing devices are assembled to simulate the physical movement in the real scene.Secondly,for scene active recognition,we design a scene active recognition system based on depth reinforcement learning algorithm.Combination of VGG network and the depth of Q-learning reinforcement learning algorithm,real scene data sets are used to train.By continuously classifying the data set and strengthening the learning experiment,the robot can recognize the scene automatically.Next,after the scene recognition of mobile robot,it needs to reach the target point according to the instruction requirements and complete the path planning.In order to make robot path planning more widely used in indoor environment,the paper puts forward the combination of the top view image and depth reinforcement learning algorithm in the environment,and directly carries out obstacle avoidance and path planning.Q-CNN network framework is designed,and the fitting Q function algorithm and the convolutional neural network are used to solve the problem of high dimension of continuous state space input data.Q-CNN network framework is designed,and fitting Q function algorithm and the convolutional neural network are used to solve the problem of high dimension of continuous state space input data.Images captured by the camera installed on the roof are input into the network framework,and robot motion instructions are directly obtained.Finally,compared with the results of the SLAM algorithm,the proposed algorithm can make the robot complete the obstacle avoidance path planning task conveniently and effectively.