This project focuses on the onboard pose estimation and trajectory generation for the quadcopter platform. The student is required to develop a suitable position and orientation estimation algorithm that is able to accurately locate the quadcopter in free space. In addition, the project will focus on synthesizing a trajectory generation routine appropriate for a quadcopter that is able to perform high-level tasks, such as waypoint following. The proposed solutions will initially be implemented in a simulation environment and ratified prior to its execution on the physical quadcopter. An off-the-shelf quadcopter has been purchased as a ready-to-go research platform.
Mathematical modelling, state estimation, control design, MATLAB and Simulink.
1x MSc (Eng), 1x Journal/Conference paper in a leading publication.
Alyssa Ramwell
Arnold Pretorius and Natasha Botha
CSIR Robotics Grant
Have successfully integrated checkerboard into the Simulink3D simulation. Can now run pose estimation on data from the simulation, and have ground truth values from which to compare. Specifically, I have been running Grunert’s P3P (Perspective-3-Point) algorithm (and the built-in MATLAB PnP algorithm) on video recorded from the “ideal” camera in simulation. After sorting out some reference frame issues (the Unreal Engine uses a left hand convention for some reason?), the results are looking promising.
