2022 IB Diploma Extended Essays

Causing the robot to correct the path creating these smaller gradients. However, the graph shows that Hector-SLAM and the Extended Kalman filter have the same changes in yaw, therefore showing that they are both robust at autonomous navigation. Similarly ORB-SLAM and ZedFu present similar change in yaw with small deviations from Hector-SLAM. With the data from the magnetometer showing the most error. Conclusion This investigation has found that the while the Extended Kalman Filter is the most robust at autonomous navigation. It isn’t generally better than Passive SLAM Algorithms such as Hector SLAM. This is shown through graphs as they provide pose values for the robot from the different algorithms. It can further be showcased that Hector-SLAM provides accurate pose estimates and as such creates an accurate path, with little deviations from that path. In comparison ORB-SLAM provides less accurate pose values with major deviations from the path, However it can still be considered a robust system for navigation as it follows the same general shape, and completes the loop in less time. Zedfu, provides accurate pose values, with some deviations from the trajectory. Again it can still be considered a robust system for navigation, as it has similar pose values and utalises an active depth sensor from the camera. The Extended Kalman Filter proves itself to be the most robust algorithm from this data-set. However is closely matched with Hector-SLAM. Limitations The experiment itself could be considered an limitation, as it meant to copy the use case of the robot in real-life scenario's most notably self-autonomous cars. The experiment fails to evaluate the SLAM algorithms in different lighting conditions, geometry and even more so fails to construct a exact an replica of the mechanisation of a car. Another major limitation is that this investigation’s data couldn’t be compared to canonical sources. Because firstly, there isn’t many available due most being closed source. Secondly, I have created my own benchmarks for evaluating the data and as such my data can’t be directly compared to a canonical source. Bias There is bias within this Investigation, due to my constructed benchmarks for evaluating the success of these SLAM algorithms. Therefore the data can also be considered a bias. The bias comes from my own improvisation of the evaluation. Based on my own knowledge, which came from canonical sources. Further Investigations A more comprehensible experiment could be done by evaluating the success of these algorithms in an outdoor environment. Having different lighting conditions, such as shadows, extreme brightness etc. and complex geometries such as mesh fences, movement and more. This would the evaluation of the algorithms in more life-like scenarios.