Accurate SLAM With Application For Aerial Path Planning

Thumbnail Image
Publication or External Link
Friedman, Chen
Chopra, Inderjit
Rand, Omri
This thesis focuses on operation of Micro Aerial Vehicles (MAVs), in previously unexplored, GPS-denied environments. For this purpose, a refined Simultaneous Localization And Mapping (SLAM) algorithm using a laser range scanner is developed, capable of producing a map of the traversed environment, and estimating the position of the MAV within the evolving map. The algorithm's accuracy is quantitatively assessed using several dedicated metrics, showing significant advantages over current methods. Repeatability and robustness are shown using a set of 12 repeated experiments in a benchmark scenario. The SLAM algorithm is primarily based on an innovative scan matching approach, dubbed Perimeter Based Polar Scan Matching (PB-PSM), which introduces a maximum overlap term to the cost function. This term, along with a tailored cost minimization technique, are found to yield highly accurate solutions for scan matching pairs of range scans. The algorithm is extensively tested on both ground and aerial platforms, in indoor as well as outdoor scenarios, using both in-house and previously published datasets, utilizing several different laser scanners. The SLAM algorithm is then coupled with a global A* path planner, and applied on a single rotor helicopter, performing targeted flight missions using a pilot-in-the- loop implementation. Targeted flight is defined as navigating to a goal position, defined by relative distance from a known initial position. It differs from the more common task of mapping, as it may not rely on loop closure opportunities to smooth out errors and optimize the generated map. Therefore, the importance of position estimates accuracy increases dramatically. The complete algorithm is then used for targeted flight experiments with a pilot in the loop. The algorithm presents the pilot with nothing but heading information. In order to further prevent the pilot from interfering with the obstacle avoidance task, the evolving map and position are not shown to the human pilot. Furthermore, the scenario is introduced with artificial (invisible) obstacles, apparent only to the path planner. The pilot therefore has to adhere to the path planner directions in order to reach the goal while avoiding all obstacles. The resulting paths show the helicopter successfully avoid both real and artificial obstacles, while following the planned path to the goal.