Scene Reconstruction (BEV) for Vision-based End-to-End Autonomous Driving with Visual Geometry Grounded Transformer

Institut
Professur für autonome Fahrzeugsysteme
Typ
Semesterarbeit / Masterarbeit /
Inhalt
 
Beschreibung

Modern autonomous vehicles leverage end-to-end learning to map camera inputs to driving decisions. However, these systems often lack explicit 3D scene understanding, such as road layouts and obstacles, limiting interpretability and robustness. To address this, we propose integrating real-time 3D scene reconstruction using the Visual Geometry Grounded Transformer (VGGT) to generate Bird’s Eye View (BEV) representations for enhanced vision-based autonomous driving.

At our research lab, we are developing a vision-based driving policy that predicts trajectories while reconstructing 3D scenes (e.g., drivable areas, obstacles, road structures) from multi-camera inputs. These BEV representations are dynamically updated, improving situational awareness and driving reliability.
 

Project Objective:
This project aims to integrate 3D scene reconstruction into end-to-end autonomous driving. Core components include:

  • Applying VGGT to reconstruct 3D scenes from camera inputs
  • Generating BEV maps for road and obstacle representation
  • Fusing BEV outputs with trajectory prediction for robust planning

We offer:

  • Cutting-edge research in perception and autonomous driving
  • Integration with an existing end-to-end driving framework
  • Opportunity to publish at top-tier AI/robotics venues
  • English-speaking environment, thesis supervision possible

Your qualifications:

  • Strong Python and PyTorch skills
  • Interest in autonomous driving and deep learning
  • Background in computer vision or 3D reconstruction is a plus

Start date is flexible. If interested, please send:

  1. A 200-word description of your computer science (e.g., deep learning, coding) course performance (campus or online)
  2. Academic transcript (optional for now, may be required later)
  3. CV (optional)

to dingrui.wangtum.de.

Möglicher Beginn
sofort
Kontakt
Dingrui Wang
dingrui.wangtum.de