PhD - Neural 3D Scene Representations for the Automotive Cabin

In this PhD project at Bosch Research, you will advance neural 3D scene representations to understand occupants and their movements from limited viewpoints inside vehicle cabins. You will integrate interior-exterior scene understanding to support future ADAS systems, collaborating with academia and industry to translate fundamental computer vision research into safety-critical applications. You will leverage known cabin geometry as supervision and modern 3D-aware priors to reconstruct occupants from one to three cameras. This role blends foundational research with practical engineering to enhance vehicle safety and comfort.

• Develop and adapt neural 3D scene representations for in-cabin perception (e.g., NeRF, Gaussian splatting, implicit/occupancy models)
• Reconstruct occupants from 1–3 fixed camera viewpoints with robust performance
• Create a unified interior-exterior scene representation to enable end-to-end ADAS training and validation
• Collaborate with academia and industry experts to bridge research and real-world applications
• Contribute to 3D computer vision and generative modeling research with a practical safety focus

• Master in Computer Science (or comparable field)
• experience in machine learning using PyTorch with vision-based models and neural scene representations
• strong Python programming skills
• solid foundations in computer vision and 3D geometry (camera models, projective geometry, multi-view constraints)
• familiarity with 3D scene representations (e.g., NeRF, Gaussian splatting, implicit/occupancy networks) or depth/3D estimation methods is strongly preferred
• prior exposure to differentiable rendering or articulated human pose estimation is a plus
• proficient English (written and spoken)
• proactive and motivated mindset with ability to navigate complex technical environments

• proactive
• high motivation
• ability to navigate complex technical environments
• PyTorch
• vision-based models
• neural scene representations (NeRF, Gaussian splatting, implicit/occupancy networks)