Uav motion planning and exploration using onboard sensors

Resum

Recently, the aerial robot is widely applied in an increasing number of civilian fields such as industrial inspection, precision agriculture, and search and rescue tasks. In order to help aerial robots to achieve autonomous missions, this dissertation presents three novel dynamic collision avoidance strategies using different onboard sensors, one autonomous navigation algorithm, and one autonomous exploration method using the RGB-D sensor respectively. Collision avoidance plays a crucial role for autonomous missions in unknown dynamic environments and still remains an ongoing research problem. When operating in dynamic environments, collision avoidance needs fast and robust algorithms for localization, controlling, and planning. Three dynamic collision avoidance strategies, such as a monocular camera-based collision avoidance method, a Light Detection and Ranging (LiDAR)-based collision avoidance strategy, and an RGB-D camera-based collision avoidance approach are presented in chapter 3. The monocular camera-based collision avoidance algorithm uses a deep learning-based obstacle detector, an Iterative Perspective-n-Point (PnP)-based pose estimation algorithm, a Receding Horizon-Covariant Hamiltonian Optimization Motion Planner (RH-CHOMP)-based trajectory planner, and a Model Predictive Control (MPC)- based controller. The LiDAR-based collision avoidance and the RGB-D camera-based collision avoidance are based on the same framework. The framework uses a Euclidean Distance Field (EDF)-based collision detecting, an EDF-Rapid-exploring Random Tree (RRT)-based path planner, a polynomial trajectory generation, and an MPC based controller. The difference between the LiDAR-based collision avoidance and the RGB-D camera-based collision avoidance is that the depth map from the RGB-D camera would be transformed to laser scan first. The main contributions of the proposed solution are: 1)A robust EDF-based collision checking algorithm is presented, which doesn't depend on the geometry of the obstacles. The presented method only uses a depth camera for collision avoidance. The EDF mapping is fast and can be applied in real-time. 2) The robustness of the proposed algorithm has been validated in different dynamic environments, running onboard. Autonomous navigation in clutter environments is a key research topic in the robotic research community. In order to provide an efficient solution for this topic, an RGB-D-based motion planning and autonomous navigation strategy are proposed in chapter 4. The RGB-D camera is used to provide the point clouds for EDF mapping. This strategy uses a novel RRT-based path candidate generation algorithm and an improved RH-CHOMP algorithm is introduced to generate the collision-free trajectory. The trajectory finally is sent to an MPC-based controller to track. The experimental results show the presented algorithm outperforms the-state-of-art and can help the aerial robot achieve high-speed flight. Aerial robots are widely used in search and rescue applications because of their small size and high maneuvering. However, designing an autonomous exploration algorithm is still a challenging and open task, because of the limited payload and computing resources onboard UAVs. Chapter 5 provides an autonomous exploration algorithm using RGB-D cameras. This algorithm uses point clouds from the RGB-D camera to generate the 3D occupancy grid map. The innovation parts of this autonomous exploration algorithm are 1) Kmean++-based frontier cluster and information gain-based frontier selection to speed up the exploration process. 2) A safe flight corridor generation algorithm enabling safe flight. All the algorithms mentioned above are validated in simulation experiments or/and real flight experiments. The experimental results show that the proposed algorithms can finish the collision avoidance, autonomous navigation, and autonomous exploration tasks successfully. By comparing with the state-of-art, the proposed algorithms all have good performance. The RGB-D-based collision avoidance algorithms have a higher success rate in high-speed flight. The autonomous navigation and motion planning algorithm can fly a shorter trajectory and achieve a higher success rate in four different challenging environments. Autonomous exploration can achieve both shorter time and shorter paths in three different environments and can be able to guide an aerial robot to perform autonomous exploration tasks. All in all, this dissertation can provide a good reference for Unmanned Aerial Vehicles (UAV) motion planning and exploration in challenging dynamic environments.