University of Twente Student Theses
Achieving full autonomy in aerial and physical human-robot interaction control via onboard perception algorithms relying on computer vision
Alonso Calderon, M.A. (2022) Achieving full autonomy in aerial and physical human-robot interaction control via onboard perception algorithms relying on computer vision.
|
PDF
12MB |
| Abstract: | In the context of aerial physical interaction, a branch of aerial robotics focuses on the study and the accomplishment of contact-based interaction and manipulation tasks with aerial robots. The goal of the assignment is to enhance the level of autonomy of the aerial robots developed at RAM-UT, providing them with an accurate localization system to perform a handover to a human. The handover will be performed in both indoor and outdoor scenarios in GPS-denied environments. Currently, the system relies on Motion-Capture (MoCap) state feedback for indoor localization. In order to find an optimal odometry technique, a comparison of the alternatives was conducted. Consecutively, deep research and evaluation on the conventional open-source VO/VIO alternatives were performed, discarding the machine learning implementations as these techniques cannot outperform the conventional ones for now. This study ranks appearance-based and feature-based methods depending on their convenience for our application. The insight gained from the analysis of open-source alternatives allows us to select ORBSLAM2, ORB-SLAM3 and SVO Pro as the most promising solutions to obtain localization in a resource-constraint platform like the MAV. These solutions were tested again the EuRoC datasets which were recorded from a MAV and are broadly used by the research community and against a custom dataset recorded at RAM-UT Arena using Intel Realsense camera D435i. The algorithm evaluation considered the overall performance of the algorithm to estimate the real trajectory, the maximum error achieved to perform human safety and the robustness of the implementations along with multiple runs. It was concluded that the algorithms SVO Pro stereo, ORB-SLAM3 stereo and ORB-SLAM3 VIO stereo are valid alternatives that would provide optimal performance under the tested conditions, providing both position and orientation of the MAV at each time step. |
| Item Type: | Essay (Master) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 53 electrotechnology |
| Programme: | Electrical Engineering MSc (60353) |
| Link to this item: | https://purl.utwente.nl/essays/92558 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page