Obstacle avoidance using point cloud data for mobile robot navigation

In this thesis, an obstacle avoidance approach has been proposed for autonomous navigation of the robot using point cloud data obtained from the RGB-D camera placed on a ground robot. While focusing on the obstacle avoidance approach, the system has been developed, including the global planner and controller parts required for autonomous navigation. It was decided to use point cloud data in the collision avoidance module of the system. The major disadvantage of point cloud data is that it is unorganized and contains numerous points. This increases the detection time of obstacles significantly. This thesis focuses on eliminating these negative aspects of the point cloud. The proposed method consists of preprocessing, detection, and planning steps. In the preprocessing step, the point cloud is arranged in an octree structure while data reduction is performed. Also, a second data reduction was achieved using only occupied voxel centers. In the perception step, the rule-based approach has been developed. This approach determines all possible situations and how the robot should act in these situations considering the occupied voxel centers. The planning step generates an obstacle avoidance plan using the Bezier curve or intermediate point. Since the proposed method is created in an iterative structure, it can avoid complex obstacles such as convex and concave. The proposed method is implemented in C++ programming language and tested with Pioneer3 - AT ground robot in Gazebo simulation environment using Robot Operating System (ROS). The robot can successfully avoid obstacles in concave and convex scenarios in addition to scenarios where the designed situations are tested in the perception step. In these cases, the time required for the robot to avoid obstacles and comparisons with OctoMap were shared.