AI Research Engineer - Dexterous Manipulation (Egocentric Models)

Flexion is developing the intelligence layer that will enable the next generation of humanoid robots to move from fragile prototypes to reliable real‑world systems. As an AI Research Engineer focused on dexterous manipulation, you will join a Zurich‑based team that builds physical foundation models using internet‑scale egocentric video. Your daily work will involve designing and implementing large‑scale pre‑training pipelines for egocentric video, creating multimodal models that combine visual input and language commands into robot actions, and building generative action heads with diffusion and flow‑matching methods. You will also devise algorithms that map human‑centric video representations onto the kinematic limits of 20‑plus‑degree‑of‑freedom humanoids, and apply offline reinforcement learning and high‑fidelity simulation to fine‑tune these models for robust manipulation in unstructured settings. Collaboration with researchers translating scaling‑law insights into production‑ready architectures is a core part of the role. The position requires a PhD or Master’s in robotics, machine learning, or a closely related discipline, with a strong emphasis on data‑driven manipulation, egocentric vision, or foundation models. You must have hands‑on experience with humanoid or dexterous manipulation, solid Python and PyTorch skills, and familiarity with distributed training frameworks such as FSDP or NCCL. Proven expertise in diffusion models, flow matching, transformers, and deploying learning‑based controllers on real hardware is essential, as is experience with reinforcement learning and simulators like IsaacLab or MuJoCo. The role offers competitive pay, an enhanced pension, generous holiday, relocation support, and the chance to work