Physical AI Researcher
Berlin, Germany | Full-time
You want your research to do more than produce another conference paper. Here, you'll work on making Foundation Models, including Vision-Language Models (VLMs) and Vision-Language-Action Models (VLAs), run in real time on actual devices and robots. You'll join an expanding European research team working at the intersection of Efficient AI and Physical AI, turning research into technology used in real-world products.
You'll research and develop new approaches to robot learning and model optimisation, from fine-tuning VLA policies and exploring World Models to quantisation, pruning and efficient inference. You'll also validate your work on real hardware, profiling latency and memory across GPUs, NPUs and robots. Your research will contribute directly to NetsPresso, an on-device ML framework that brings optimised AI models into customer products.
This is an opportunity to take ownership of research from identifying problems and developing new methods through to benchmarking, publication and deployment. You'll collaborate with research, framework and embedded engineering teams in Europe, alongside wider R&D teams in Korea. As an early member of the European team, you'll have real influence over the research direction.
You'll need a PhD or equivalent research experience, strong Python and PyTorch skills, and experience in at least one of model optimisation or Foundation Model training and fine-tuning. Experience with VLMs, VLAs, robot learning, on-device deployment or tools such as TensorRT and CUDA would be a real advantage. Publications at leading AI or robotics conferences are also valued.
If you want to work on emerging AI research that makes its way out of the lab and onto real devices and robots, let's have a conversation.
You'll need:
-
PhD or equivalent research experience in a relevant field.
-
Experience in model optimisation or Foundation Model development.
-
Strong Python and PyTorch skills.
-
Ability to independently research, reproduce and validate published methods.
-
Interest in taking research through to real-world deployment.
