Background:

Autonomous driving systems must perceive the current 3D scene while anticipating its future evolution. Compared with bounding boxes, semantic occupancy provides a dense representation of scene geometry and semantics, making it better suited for modeling occlusions, road structures, irregular objects, and complex dynamic environments. However, most visionbased occupancy prediction and forecasting methods remain limited to closed-set categories and struggle with rare or unseen objects. Recent works such as Cam4DOcc [1] and ForecastOcc [2] have advanced camera-only 4D occupancy forecasting and vision-based semantic occupancy forecasting from image sequences, while PG-Occ [3] explores open-vocabulary occupancy prediction with text-aligned representations and Gaussian-based scene modeling. Yet these directions remain largely separate. This thesis will study vision-based open-vocabulary semantic occupancy forecasting, aiming to predict future 3D semantic occupancy from camera observations while supporting flexible text-driven categories for more generalizable, future-aware autonomous driving scene understanding.

Your Tasks:

• Conducting a literature review on camera-only occupancy prediction, 4D occupancy forecasting, and open-vocabulary 3D scene understanding.
• Designing a model that predicts future 3D semantic occupancy from multi-view camera sequences.
• Evaluating the proposed method on autonomous driving datasets and comparing it with relevant closed-set and open-vocabulary baselines.

Your Profile:

• Strong background in computer vision, deep learning, robotics, or related fields.
• Motivation to work on open-ended research problems and conduct independent experiments. 
• Solid programming skills in Python and experience with PyTorch.
• Good analytical skills and ability to communicate research ideas clearly in English.

Start date: Immediately

Duration: As per the applicable examination regulations.

If you are interested or have any questions regarding this thesis position, feel free to contact.