Thesis List

his master’s thesis will investigate recurrent state-space models for processing asynchronous event-camera data. The focus will be on designing and evaluating architectures that can efficiently model long temporal dependencies in event streams while maintaining low latency and computational cost. The study will explore how modern state-space models can be adapted to event-based vision tasks such as object detection, object tracking, or motion understanding, and compare them with recurrent neural networks and Transformer-based approaches. The goal is to assess the suitability of state-space models for event-driven perception and identify architectural choices that improve accuracy, efficiency, and real-time performance.

Requirements:

We are looking for a motivated student who enjoys combining computer vision, programming, and hands-on experimentation. Ideally, you bring:

• Solid programming skills in Python and enthusiasm for developing and testing new ideas
• Basic knowledge of computer vision and an interest in modern vision algorithms
• Understanding of camera models, projections, and coordinate frames
• Curiosity about robotics and motivation to gain hands-on experience with real systems

This master’s thesis will conduct a Systematic Literature Review (SLR) on object detection and object tracking using event-based cameras. The review will follow the Kitchenham methodology for systematic literature reviews and the PRISMA framework for transparent study selection and reporting. The study will analyze existing methods, event representations, neural network architectures, datasets, evaluation metrics, and performance measures used in event-camera-based object detection and tracking. The goal is to summarize the current state of research, compare existing approaches, identify limitations in current evaluation practices, and highlight open research challenges and future directions in event-based computer vision.

Deep learning models for 3D LiDAR perception generally require large volumes of manually annotated point clouds, and 3D bounding-box annotation is expensive, slow, and error-prone. As a result, self-supervised learning (SSL) is of interest to address this bottleneck by pretraining general-purpose point representations on unlabeled data, which can then be fine-tuned on small labelled sets. Most recent progress in this direction — including the Sonata architecture and its extension Vernata — has been developed and evaluated almost exclusively on outdoor, automotive-grade LiDAR data, using dense, high-channel-count sensors paired with RGB cameras. It is interesting to see if extending the method to logistics and warehouse robotics datasets with a different sensing regime: confined indoor spaces, sparse low-channel-count sensors, a small and fixed set of object classes (pallets, forklifts, transport platforms, boxes), and deployments that frequently lack a co-registered RGB camera, is feasible. Thus, this thesis proposes to reproduceVernata, a recent (IROS 2026) multi-modal, multi-teacher self-supervised distillation framework for LiDAR point clouds, and to systematically evaluate and adapt it for 3D object detection on the ANavS/KIT warehouse LiDAR dataset.

The project involves the design and implementation of an actuated stabilization platform, inspired by a Stewart platform, mounted on a mobile robotic base to maintain a payload in a stable inertial pose during aggressive maneuvers. The student will work on mechanical design, embedded systems, sensor integration, attitude estimation using IMU and Vicon motion capture, and feedback control development, with extensive experimental validation in a robotics laboratory. 

This thesis offers hands-on experience in advanced robotics, control systems, and motion-capture technology, and is ideal for students with a strong interest in mechatronics, robotics, control engineering, or autonomous systems. Applicants with experience in CAD, programming (C++/Python), ROS, or control theory are especially encouraged to apply.

You may find more detailed information here.