Research On AGV Navigation And Motion Planning Method Based On Vision

With the progress of science and technology,intelligent robots gradually play an important role in our lives.Robots are widely used in logistics,warehouses,docks,intelligent garage and so on.In recent years,people concerned about the parking system research,and intelligent garage has gradually become a hot research.There are many ways of navigation,such as magnetic tape navigation,magnetic nail navigation,inertial navigation,laser navigation and so on.These navigation methods can complete the relevant task requirements,but there are some obvious shortcomings.When using magnetic tape and magnetic nails,we need to decorate the scene.The application is cumbersome and the cost is also high.Inertial sensors in inertial navigation drift and the cumulative error increases with time.The reflector board approach is a widely used laser navigation method.But the road signs in the garage are easily blocked by the vehicle and the laser scanner prices are relatively high.These traditional positioning modes do not have the ability to relocate,and the overall method is lacking in robustness.In order to solve the above problems,this thesis designs a navigation method that uses realtime visual SLAM.The navigation method is built on the heavy-duty AGV and we do the experiments.Aiming at the application requirements,we design the overall navigation scheme.The navigation scheme is divided by function.The navigation system is designed,which is divided into human-computer interaction layer,navigation layer,motion control layer.The camera model and AGV mathematical model are established,we carry out the kinematic analysis of AGV.The marker is used as a visual landmark in the navigation method.And we can get the pose of camera in the marker coordinate system to correct AGV pose.We use the tracking method to ensure the real-time calculation of the pose in the SLAM front-end process.The system uses the local map to achieve pose optimization,which is to ensure the robustness and the accuracy of the results.The key frames are filtered in the mapping thread and the pose of the key frames and map points is optimized again,which is to ensure the sparseness and correctness.And we use Bo W to increase the relocation function and closed-loop detection.The SLAM tested with the public dataset has a good performance.The sparse point cloud map does not have the geometric characteristics of the scene,and it can't be used for path planning.We build the global grid map according to the actual scene and use the A-star algorithm for path planning on it.The system obtains the transformation matrix according to the pose of the current frame in the marker coordinate system during SLAM initialization.We use the Backstepping algorithm to design the trajectory correction control law based on real-time location information and ideal trajectory to solve the trajectory offset problem.And the related simulation is carried out.We set up the navigation method in the experimental prototype and complete the verification of the overall function.