Research On Simutaneous Mapping And Automated Navigation For Photovotanic Cleanning Robot With Kinect2

As the increasing use of photovatanic system, it becomes a hard problem to keep them clean. Current cleaning solutions including manual cleaning, spray system, on-board robot and cleaning cars have their weaknesses and cannot be widely applied. More study need to be done to make cleaning robot more intelligent and cheap.This project is aimed to develop a cleaning robot for photovotanic plant to complete the cleaning task in half-structured environments without human interference. Equipped with Kinect2, the robot must map the 3D environments in a robust and real-time way, navigate to a certain position with the map it built with Kinect2, and clean the photovatanic panel.Four aspects of work have been done as follows:(1)Kinect2 calibration and point-line feature extraction.The measurement errors and their sources are analyzed and tested, thus along with the improvement of using conditions, Kinect2 calibration experiment is also carried out to decrease the measurement errors. To get more useful information about the working environments of photovotanic cleaning robot, the project also study on how to extract point features and line-segment features more robust and quickly.(2)Real time 3D mapping with points and line segments.Reprojection errors for point features and line-segment feature are defined in 3D space, and the 3D pose transformation between two raw frame data can be got by minimizing the sum of reprojection errors, resulting to the registration of two raw point clouds. And to get redundant constraint for closed loop, a loop detection approach is provided based point and line-segment feature together. General graph optimization is performed with Levenberg–Marquardt algorithm to decrease the accumulated error. And lastly, the point clouds are transformed to Octomap because it’s complicated and hard to maintain. The Octomap is compact, easy to maintain and more importantly, it can be directly applied to robotic navigation and path planning.(3)Autonomous navigation with D* and Elastic Band method.With the Octomap built with real time 3D mapping, the robot can navigate to a certain position as appointed. By combining two mature algorithms, the robot can tackle with dynamic environments, modify current path solution gradually, and get the optimal solution finally. Besides, this project also takes the nonholonomic constraints into consideration; elastic band algorithm is used to modify the local path based on the global path planned with D* algorithm.(4)Experiments for robotic simultaneous 3D mapping and navigation.Gazebo simulations with ROS and real robotic experiments are carried out to check the robustness of the project. The real time 3D mapping algorithm and path planning algorithm are tested in many different scenarios, including indoor and outdoor, the correctness and robustness are verified for photovotanic cleaning robot. What is more important is that the simulation and real robotic experiments also show the algorithms can also be applied to many other scenarios.