| At present,improving the intelligence and autonomy of mobile robots has become the direction of robot development.The control of mobile robots based on vision and deep learning combines robotics,image processing,mechanical electronics,communication technology and machine learning,and has become a hot and difficult point in the current research.In view of the problem of insufficient expressive ability and combination flexibility of existing behavior models,a hierarchical behavior model was designed,and the concept of behavior tree was introduced to develop and improve the existing behavior model.In addition,by adding a dynamic scheduling node to give the dynamic scheduling capability to the behavior tree,it solves the problem that the traditional behavior tree can hardly express non-logical behavior decisions.For the problem that deep learning is difficult to be applied in the mobile robot behavior control due to its dense computation,a lightweight convolutional neural network model was designed based on the decomposable convolution,and the model was further accelerated by video stream resampling.The availability of the model in the process of behavior control was verified by experiments.Aiming at the problem of poor autonomy of current mobile robots,an autonomous behavior control algorithm according to the behavior tree was designed based on hierarchical behavior model and object detection,and the feasibility of the algorithm was tested through experiments.A behavior learning algorithm based on random forest was proposed and the learning of behavior was realized.The effectiveness of the algorithm was verified by experiments.The hardware system based on Arduino Uno controller was built,and multithread programs were designed for local control and remote decision making of the robot.The related algorithms were tested experimentally,and the object detection and positioning of the mobile robot and the robot navigation control based on visual feedback were realized. |
PDF Full Text Request