Research And Application Of Local Vision-based Navigation System

In modern life,mobile robots have broad application prospects,and irreplaceable role in some fields.Therefore,this field has attracted a large number of outstanding scientists and funds.This kind of robot has been initially applied to space exploration,underwater exploration,automated warehouses,service robots and so on.Autonomous navigation system is the core of mobile robots.The system needs to face the real environment alone without manual intervention and guide the robot hardware system to arrive destination.Since the natural environment is usually complex and changeable,there is no set of autonomous navigation algorithms that can adapt well to various problems.Different navigation systems are usually designed based on the common navigation system framework and specific application scenarios.In this thesis,the application scenario is to achieve material transport by mobile in large-scale scenarios.The characteristics of the task are: The map of material transport is relatively large;the starting point and end point of material transportation are relatively fixed and the period of variance is long.According to the characteristics of application scenario,based on the analysis of existing navigation methods,this article believes that our system does not need to complete high-level identification and map construction steps which the general-purpose navigation system need,just scene identification step.And scene recognition step can be quickly implemented by local visual features of the scenario.Based on the above analysis,this thesis uses local vision as basis to design our navigation system.The system does not require establish a complete environmental map,it only requires a manual demonstration of navigation process.The navigation system will then automatically mimic the exemplary navigation trajectory and avoid obstacles if necessary.In this thesis,we mainly carried out the following work:1)Based on recent image processing technology and deep learning technology,a global path planner suitable for the application environment of this paper is implemented.it not only avoids construct global maps and high-level recognition,but also own a relatively simple training process.2)By studying local path planner module and local cost map module of laser navigation system,we designed the local path planner module and local cost map module of our navigation system.3)By studying the framework of laser navigation system,we designed the framework of our navigation system.In the system design,the system stability and the dynamic pluggability of the module are taken into consideration.4)Based on the physical environment simulator gazebo,we designed a simulation experiment,and verify the performance of our system.