Life is always about becoming... In life it is about going on a journey to become the best version of our future self. As we discover new things, we face challenges, master them and grow beyond us.

Apply to Mercedes-Benz and find the area where you can develop your talents individually. You will be supported by visionary colleagues who share your pioneering spirit. Joining us means becoming part of a global team whose goal is to build the most desirable cars in the world. Together for excellence.

Number: MER00046D1 Mercedes-Benz Group AG is one of the most successful automotive companies in the world. With Mercedes-Benz AG, the vehicle manufacturer is one of the largest providers of premium and luxury cars and vans. In our cross-functional team AI Research – Physical AI, we explore the latest technologies for AI-based robotics architectures and modern AI stacks. A central research field is the learning of complex tampering skills from human demonstrations. In this master's thesis, you will explore how teleoperation, demonstration data and vision lane action models can be used to develop powerful robot learning systems. Dextere robot hands enable the execution of complex manipulation tasks and are considered to be an important field of application of modern robot learning methods. Current Learning-from-Demonstration and Vision-Language Action approaches promise an efficient transfer of human capabilities to robot systems. However, it is open to what factors influence the learning performance and generalisation capability of these procedures. The aim of the work is the scientific investigation of the relationship between demonstration data, teleoperation procedures and the performance of modern robot learning approaches. For this purpose, demonstration data with a dexter robot hand are collected, various learning methods are trained and their generalisation capability is systematically evaluated under different experimental conditions. Possible research questions are: What influence does the quality of hand-retargeting have on the performance of learned manipulation strategies? How does the size and diversity of the demonstration data set have an impact on success rate and generalisation capability? What influence do different camera configurations and sensor systems have on the learning performance? To what extent do Vision Language Action models benefit from task-specific fine tuning compared to classic imitation learning? What factors limit the portability of teleoperated demonstrations to unknown objects and manipulation scenarios? The work will provide new scientific findings on the relationship between teleoperation, demonstration data quality and the performance of modern robot learning processes. The results are intended to contribute to a better understanding of how dextere manipulation capabilities can be efficiently transferred from people to robotic systems using learning-from-demonstration approaches and vision-nguage action models. In addition, concrete recommendations regarding data acquisition, demonstration quality, sensor technology and model training are to be derived for future robotic learning systems. These challenges come to you: Literature research on dexterer teleoperation, hand-retargeting, imitation learning as well as modern vision language action models (e.g. ACT, diffusion policy, OpenVLA or π0) Construction or extension of a teleoperation platform for controlling a five-five robot hand Development of a hand-retargeting pipeline for transferring human hand movements to the kinematics of a dexter robotic hand while maintaining the intention of manipulation Definition of representative manipulation tasks and the establishment of a synchronized data acquisition pipeline for the integration of multi-camera image data, proprietary signals and robot actions Collection, preparation and analysis of a demonstration data set for learning-from-demonstration approaches Training and benchmarking of imitation learning baselines as well as fine tuning of modern vision-Language action models on the collected data Planning and implementation of controlled experiments to investigate the influence of demonstration quality, quantity of data, record diversity and sensor technology on model performance Scientific evaluation of the resulting strategies on a real robot platform with regard to success rate, robustness and generalization on unknown objects, object positions and manipulation situations Analysis of results and derivation of scientifically sound recommendations for efficient learning-from-demonstration systems The activity can start from October 2026.

Current Master's degree in Computer Science, Robotics, Artificial Intelligence, Electrical Engineering or a comparable course of studies

Good python skills and first experiences with machine learning or robotics Interest in Robot Learning, Computer Vision and Embodied AI Ideally practical experience with deep learning frameworks (e.g. PyTorch) and Linux Self-employed, structured and scientific work Secure knowledge of German and English in word and writing

Commitment and team skills

Additional information: We are looking forward to your online application with CV, lettering, certificates, current enrollment certificate, giving the semester and proof of the regular study period. Please do not forget to mark your documents as ''relevant for this application' in the online form and to observe the maximum file size of 5 MB. Further information on the setting criteria can be found here. Disabled and equalized applicants are welcome! The severely disabled representative (sbv-sindelfingen@mercedes-benz.com) is happy to support you in the application process. People Solutions will be happy to help you with questions about the application process. You can reach us by email via myhrservice@mercedes-benz.com or by phone at 0711/17-99000 (Mo-Fr 10-12am & 13-15am).

Food supplements Employee handy possible Employee discounts possible Employee participation possible Staff Events Coaching Flexible working time possible Hybrid work possible Health measures Employment Mobility offers Parking space Business doctor Good connection Accessibility Child care Kantine, Café