2022 IB Diploma Extended Essays

rough shape as the LiDAR. This would suggest that there is significant scale drift from both the visual slam systems, which could be caused due to loop closure problems. While the LiDAR doesn’t suffer from scale drift as it uses an active depth sensor (laser). As such calculating pose for the system becomes increasingly easy. The Extended Kalman filter presents little to no error within the first graph. While it keeps the general shape of the curves from the passive slam algorithms, it has a much smoother gradient and generally doesn’t need to correct itself as much, it also takes sharper turns than the hector slam and zed-fu. This graph therefore shows that EKF is much more

robust at autonomous navigation than passive algorithms. Graph 2. Changes in y coordinates in relation to time

Graph 2. Illustrates the changes in y coordinates (m) in relation to time (seconds) for Hector SLAM, ORB-SLAM, Zedfu and Extended Kalman filter. Within the graph the min and max highlight the robot changing y coordinates in order to make a turn. Therefore, changing direction. The smaller changes within the curve highlight the robot correcting itself within the y plane to follow the path. The graph demonstrates again that LiDAR is the most robust passive SLAM algorithm. Having no errors and presenting a very smooth gradient. In contrast, ORB-SLAM presents errors within it’s pose. As it having to make slight corrections in relation to the y plane. Though it follows the same general shape as the LiDAR. Similarly the Zedfu algorthim presents little to no errors, and generally follows the shape LiDAR, though it has higher maximum's and