3 open thesis projects

Multi-drone trajectory optimization for cable manipulation

Brief description: In this project we will examine the feasibility of using drones for repairing cut power or communication cables suspended from poles. We will examine a scaled down mock-up of the problem, where two small indoor drones will have to pick up and manipulate a rope or light cable from floor level to a suitable mock-up power pole. The precise methods used, and ambition will be adapted based on the number of students, their interests and skills. It is possible to focus the project one perception, trajectory optimization and coordination, or robot learning; however, in all cases, the goal is to be able to demonstrate task completion in the mock-up lab scenario (with potential simplifications to sub-systems). Skills developed: Robot learning, trajectory optimisation, development with hardware in the loop 

Lend me an eye: active multi-view diffusion policies 

Brief description: Diffusion policies are a promising approach to represent policies learned from demonstration. Often these policies operate directly based on camera image input, relating incoming images to a sequence of robot end effector commands. While this works well for simple tasks, complex tasks often involve a high degree of occlusion, where the robot or objects in the scene can obscure the view of important task-relevant variables. In this project, you will explore the idea of using a movable camera to observe the relevant parts of the scene and training a second policy that moves the camera view to improve the success rate of the learned diffusion policy. You will explore learning these policies jointly, through demonstration or through reinforcement learning. Skills developed: Robot learning

Multi-sensor localization fusion for an autonomous tractor

Brief description: In this thesis you will develop a framework for fusing multiple sources of localization data for an autonomous tractor. You will primarily be focused on interfacing and processing data coming from a global navigation satellite system (GNSS). In addition, you will develop a method for calculating odometry from wheel encoder sensors, as well as IMU and visual odometry data. The main task of the thesis will then be to combine these various sources of data in a filtering framework to obtain smooth and stable localization estimates.