Intelligent Mobile Robot System Based On Spiking Neural Network And Object Detection

In recent years,with the continuous development of science and technology,mobile robots are not only limited to industrial use,but gradually enter and integrate into the lives of ordinary people.The broad application scenarios and huge development potential of mobile robots make them a hot spot of current research.Under the current boom of deep learning,the exploration of the combination of mobile robots and deep learning is a current mainstream.After model iteration and extensive training,deep learning networks can often cooperate with mobile robots in many aspects,such as road pedestrian detection,security robots,etc.However deep learning also causes problems of high computational consumption and high energy consumption.For mobile robots,especially robots with specific usage environments,there are strict restrictions on computing power and power consumption.In the current era when deep learning and mobile robots have been closely combined,this thesis attempts to combine deep learning and mobile robots with more novel,lower energy consumption,and better dynamic network structures.The Spiking Neural Network is known as a new generation of artificial neural network,which is a network structure more similar to the real brain in the underlying mechanism,so it also has the advantages similar to the brain: high dynamics and low power consumption.Based on the above research background,this thesis mainly studies the use of Spiking Neural Network and traditional deep learning in mobile robots,including Spiking Neural Network control module and deep learning object detection module.The main contents include:(1)The supervised learning algorithm in the Spiking Neural Network is used to regulate the wheel speed of the two-wheel differential drive trolley,enabling it to solve the trolley yaw problem caused by the difference in driving characteristics of the wheel motors themselves,the difference in assembly accuracy and the disturbance of the external ground such as bump.The Spiking-Neural-Network-based control algorithm enables the trolley to move with an angular deviation of no more than 2,similar to the performance of PID,and is able to adapt to diverse usage environments by replacing the learning rules.(2)Combine the continuous attractor network to encode the orientation information of the car.Through the interference test of the attractor network,it is verified that the attractor network has strong robustness and good resistance to interference such as noise and outliers.(3)A deep learning-based target detection algorithm is used to detect the target.The object detection algorithm is trained using the data-enhanced mask dataset and obtains a robust effect.Combining the Spiking Neural Network part with the deep learning object detection,a mobile robot system is constructed,the tracking task experiment of the object without a mask is carried out,and a map is generated using the following characteristics of the car.In this thesis,the algorithm based on Spiking Neural Network and the control of unmanned vehicles are combined,and good results are obtained,and the Spiking Neural Network is combined with the object detection based on deep learning to verify the cooperability of the two,and provide low-cost mobile Robots offer new system ideas.