Doctoral Students

Master Students

Synthetic Images for Convolutional Neural Networks in Autonomous Drone Racing

In this work, the efficiency and robustness of tailored dataset generation for deep learning-based computer vision algorithms, for the specific case of autonomous drone racing, is analyzed and discussed. This challenge is a perfect opportunity for researchers to stretch the limits of autonomous robotic systems, since it combines the sensing, planning and acting phases into one complex problem. The proposed solution focuses on the perception of self-flying drones using state-of-the-art convolutional neural networks, and also attempts to solve the sensing and planning problems as a proof-of-concept for the vision algorithm. Effectively, this research concentrates on developing a robust and generalized vision algorithm for detecting obstacles regardless of their shape and color. The end goal is to greatly facilitate the learning of the vision-based detection model, by automating the tedious and long process of collecting and annotating a large dataset required to train the object detection algorithm using computer graphics. Indeed, this task is often the most time-consuming part of training convolutional neural networks, mostly because each image must be annotated to be used in supervised learning. Finally, experiments are conducted and promising results are shown, which promotes this original idea of using synthetic images to train convolutional neural networks for drone racing. (video)

Bachelor Students

Design, Manufacturing and Control of a Gripper for Unmanned Aerial Vehicle

In this work, the objective is to develop a versatile and low-cost gripper that is easy to manufacture and capable of grasping objects. After reviewing and evaluating of past researches, it was decided that the prototype for this project will be a 3-degrees of freedom gripper which will have elevation, yaw and also grasping end-effector. The whole prototype was manufactured by utilizing 3D printing and then assembled together manually. The prototype was then put through a series of static and flight tests to capture common household objects in order to evaluate its performance. The results of the experiments showed that the gripper is able to consistently capture the objects effectively under different circumstances. The prototype also took advantage of the semi-autonomous control during the experiment to enhance its capability of autonomous grasping. (video)

Design, Manufacturing and Control of a Small-Size Unmanned Aerial Vehicle

In this work, the goal is to design and produce a prototype of a small-sized Y6-shaped coaxial hexarotor. The Y6 prototype would be constrained to having a motor to motor distance of lesser than 25cm. Multi-rotors accessories available in the market would be selected with this design constraint in mind. In addition, the Y6 would be equipped with additional modules to perform autonomous flight. The frame of the Y6 would be designed based on the selected multi-rotor accessories using Computer Aided Drawing (CAD) software while referencing to previous models of Y6 that had been designed. The frame was manufactured using the 3D printing technologies. The Y6 was controlled initially with low-level controllers available in the market. Once satisfactory amount of stability has been achieved, high-level controllers was implemented on the Y6 to achieve autonomous trajectory flight. At the end of the project, the designed prototype was capable of achieving both manual and autonomous flight. (video)