Collaborative Aerial-Ground Multi-Modal Autonomous Navigation

In recent years,many intelligent robotic applications have increasing demands for full automation of robotic systems,including mobile robots that must reach the specified target locations or areas to perform tasks autonomously without prior knowledge.Aiming at the problems of weak environmental perception and poor self-localization accuracy of a single mobile robot,this thesis proposes a cooperative aerial-ground multi-modal autonomous navigation method to enhance the self-localization and mapping capabilities of the robot system.At the same time,the system autonomously completes the path search for target locations,which guide the ground robot to reach the target location safely and stably.In order to achieve collaborative aerial-ground multi-modal autonomous navigation,this thesis divides the autonomous navigation method into three sub-modules.Research on each sub-module,the main work and contributions are as follows:1.A collaborative aerial-ground multi-modal localization and mapping method is proposed,which allows aerial-ground robots to jointly establish the perception in complex unknown environments.This method obtains aerial-ground transformation relationship and real-scale information by detecting and identifying visual markers.Therefore,the drone can take advantage of its unique aerial field of view to provide sufficient obstacles and terrain information for the ground end.On the contrary,the ground robot optimized the aerial end's pure visual pose and its corresponding map points with the stable multi-modal fusion localization information.In the end,the respective local maps of aerial-ground robots will be fused into a consistent global map.Outdoor experiments show that this method effectively increases the environmental perception of the robot system.Compared with a single robot,the average translation accuracy of aerial-ground estimated trajectory is increased by 0.099 meters,which significantly reduces the cumulative error of localization and mapping.2.This paper presents a map representation and path planning method for cooperative aerial-ground system.In order to give global navigation information to the ground robot,this method constructs an octomap with the aerial-ground fused 3D point cloud which is the output of the cooperative aerial-ground multi-modal localization and mapping,and then maps the octomap into a 2D occupied grid map which is suitable for path planning at the ground end.By implementing the improved A* path planning algorithm in this paper,the optimal global path to the set target point in the 2D map is obtained.By running the systemic test,this global path can prevent the ground robot from falling into a local dilemma.3.This paper proposes a dynamic trend-aware ground robot trajectory planning method to help ground robots cope with dynamic obstacles in complex and unknown environments.By predicting the trend trajectory of dynamic obstacles,the method can achieve the real-time evasive maneuver of the ground robot to avoid the dynamic obstacle during the autonomous navigation process.Even if the detection state of the dynamic obstacle is unstable,the method can ensure that the generated maneuver trajectories are safe and reliable.The trajectories satisfy the constraints of the kinematics,dynamics and static obstacles of the ground robot.Simulation experiments show that in a dynamic environment,the speed stability of the ground robot using this method is increased by 82.75%,the average path length is shortened by 8.42%,and the average time consumption is reduced by 19.29%.Based on the above work,a complete cooperative aerial-ground multi-modal autonomous navigation system is constructed in this paper.The actual operation test of the system shows that the system can stably and robustly complete the autonomous navigation task.