[Hiring] Internship / Master Thesis - VLAMs for Trajectory Prediction @Volkswagen AG

Role Description

We offer you an exciting opportunity for your master thesis or for an internship in our ADAS/AD – VLAM team in the field of model engineering for autonomous driving. For the following topic you get the responsibility: multimodal trajectory prediction with Visual-Language-Action models (VLAMs). The aim of this master thesis / the internship is to explore how large, pretrained multimodal foundation models can:

• Provide rich scene and intent representation
• Represent latent and uncertain dynamics
• Generate accurate, diverse, uncertainty-aware agent trajectory forecasts that can be used for downstream vehicle control

The department works on software and machine learning models for automated driving (AD) in urban environments, with a focus on the use of foundation models for end-to-end training. Within this department, we are a team of ambitious and highly motivated experts in the field of self-driving vehicles working in an agile environment to advance the autonomous driving stack.

Qualifications

• Master student in the area of Computer Science, Robotics, Mechatronics Engineering, Electrical Engineering or similar
• Good general knowledge in the field of (self-)supervised learning, transformer-based architectures, and (vision) foundation models
• Very good knowledge of machine learning frameworks such as PyTorch and applied knowledge in software development and programming in Python or C++
• Structured and independent work, above-average commitment and flexibility
• Strong communication skills and analytical understanding

Requirements

• Research and assess the state-of-the-art of visual-language-action models for autonomous driving
• Develop and refine model architectures focused on outputting diverse, probabilistic, and plausible future trajectories rather than a single deterministic path
• Conduct extensive experiments on public and internal datasets
• Investigate the limitations of open-loop evaluation on standard datasets that inherently contain only a single realized future (one ground truth)
• Work closely with our PhD students and the model engineering team

Benefits

• Remote work options within Germany
• Duration: 3 - 6 months
• 35 hours/week
• Salary: 13,90 €/hour