Student Projects
I actively supervise MSc projects / internships. If you are interested in any of the following topics, or have a general idea of research related to planning / control / estimation of aerial robots, please feel free to contact me at s.sun-2@tudelft.nl.
Please also send with your application an updated CV, a transcript of your current education and indicate a planned start date.
Whole-body Planning and Control of an Aerial Robotic Manipulator
In this project, you will work on designing whole-body control of aerial manipulators without the quasi-static assumptions. You will have to directly control the torque level of the actuators and jointly account for the multiple DoF kinodynamics of both the manipulator and the base of the drone. As a result, the aerial manipulator is expected to have unprecedented agility and maneuverability compared to state-of-the-art solutions. You will also work closely with the engineer to build and modify the air manipulators in the lab.
Actuator Dynamic Compensator for Incremental Nonlinear Control
Multi-rotor drones, known for their agility, still face challenges in following highly maneuverable references or rejecting random wind disturbances due to the dynamics of their actuators. The rotor speed commands generated by the controller cannot be executed instantaneously by the motors. This limitation impacts their accuracy in outdoor windy conditions and reduces their agility in cluttered environments. In this research, the student will explore a technique known as “command advancing”. Preliminary studies have demonstrated its effectiveness in overcoming the limitations posed by first and second-order dynamic actuators. The student aims to refine and implement this technique on a BLDC (Brushless DC electric) motor and subsequently assess its performance on a quadrotor drone.
Contact-based incremental nonlinear control for aerial manipulators
Aerial manipulators are aerial robots that can manipulate the object and physically interact with the environment. In other words, they are also “flying hands” rather than “flying cameras”. An emerging application of aerial manipulators is using them for ultrasonic testing (UT) of infrastructure, such as wind turbine blades to detect their interior flaws. While existing solutions can achieve point detection, it still remains a challenge to smoothly slide on a rough surface because of the unknown frictions. In this project, you will develope and validate planning and control algorithms in this challenging task.
Privileged learning for distributed aerial robot control for collaborative object manipulations
Transporting and manipulating an object in the air with multiple aerial robots is a promising technique. However, existing approaches typically rely on a centralized policy to fully leverage the robots' capabilities. This reliance is due to the centralized controller’s ability to observe or estimate the full state, and generate control commands that comply with kinodynamic constraints of the system. Nevertheless, such a system often struggles to scale to large numbers of agents and is constrained by limited computational bandwidth. In this research, the student will investigate the potential of privileged learning to address multi-agent control problems, focusing on the use of multiple aerial robots for collaborative object manipulation in the air. An existing centralized policy based on nonlinear Model Predictive Control (MPC) will serve as the basis for the teacher policy. Experimental validation of the algorithm is anticipated.
