About the role
The AI Research Division of Agile Robots is looking for a Working Student (m/f/d) Simulation-Based Fine-Tuning & Benchmarking for VLA-Models. You will support research on simulation-based fine-tuning and benchmarking of Vision-Language-Action (VLA) models for robotic tasks. The role focuses on building reliable evaluation workflows in simulation, running controlled experiments, and improving understanding of model behaviour across tasks and settings. You will work closely with senior AI engineers on research infrastructure and experimentation for embodied AI.
Your Responsibilities
- Benchmarking: Develop and maintain simulation-based benchmarks for evaluating VLA models across robotic tasks.
- Model Integration: Implement modular interfaces between VLA models and simulation environments.
- Experiment Design: Design and run controlled experiments, including hyperparameter studies and ablations.
- Fine-Tuning: Support fine-tuning of VLA models in simulation and analyze training behaviour and learning dynamics.
- Result Tracking: Build and maintain pipelines for logging, organizing, and comparing experimental results.
- Failure Analysis: Analyze generalization gaps, failure cases, and behavioural differences across models and training setups.
Essential Skills
- Enrollment: Currently enrolled in a Master's program or an advanced Bachelor's program in Computer Science, Machine Learning, Robotics, or a related technical field.
- Python: Strong programming skills in Python.
- PyTorch: Hands-on experience with PyTorch for training and experimentation.
- ML Workflows: Good understanding of deep learning training workflows, experimental design, and model evaluation.
- Analysis: Ability to interpret experimental results and reason about model behaviour in a structured way.
- Robot Learning: Exposure to reinforcement learning, imitation learning, or robot-learning workflows.
- Problem Solving: Strong analytical thinking and a structured approach to technical problems.
Beneficial Skills
- Embodied Models: Familiarity with Vision-Language-Action models, multimodal learning, or embodied AI.
- Simulation Workflows: Experience with robotics simulation environments such as Isaac Gym, MuJoCo, or Habitat.
- Experimentation: Familiarity with hyperparameter optimization, ablation studies, or large-scale experiment workflows.
- Research Tooling: Experience with experiment tracking tools such as Weights & Biases and interest in model robustness or generalization.
What we offer
- Practical learning opportunities to complement your studies.
- Dynamic high-tech company combined with financial soundness and world class investors.
- Join an interdisciplinary, international team with 60+ different nationalities in a collaborative work environment.
- Corporate Benefits Program that covers health, mobility and learning with 100 € net per month.
- Modern office facilities with a rooftop terrace overlooking Munich, free drinks & fruits, and regular company events contribute to a good working environment.