Reinforcement Learning Research Scientist for Dexterous Manipulation

Your mission & challenges

Together, we are taking the step into a new era of cognitive robots:

• Advanced AI for humanoid robotics: Design, train, and deploy next-generation learning-based policies that enable humanoid robots to perform dexterous manipulation and coordinated whole-body behaviors in the real world.

• Foundation Models: Fine‑tuning VLA policies with deep reinforcement learning to achieve highly dexterous, simulation‑driven manipulation.

• End‑to‑end RL pipelines: Build complete reinforcement learning systems, from data generation and environment design to large‑scale training, evaluation, and deployment on physical robots.

• State-of-the-art learning methods: Advance reinforcement learning, imitation learning, and sim‑to‑real transfer to enable scalable, reliable humanoid behavior.

• Benchmark-driven quality: Design and evaluate robotic policies using modern manipulation benchmarks such as CALVIN, RoboCasa, and related large-scale test suites.

• Deep hardware collaboration: Collaborate closely with hardware and control teams to seamlessly integrate your models into real robots.

• From simulation to real robots: Validate and iterate on algorithms through real-world experiments, closed-loop testing, and full sim‑to‑real deployment.

What we can look forward to

• An excellent Master’s or PhD in Computer Science, Informatics, Robotics, Physics, or a related field

• A proven track record: Your projects, patents, and open-source or research contributions demonstrate measurable impact.

• The desire to go beyond the state of the art – you don’t just want to improve, you want to create something new.

• Strong foundation in deep reinforcement learning, imitation learning, and modern ML architectures

• Experience developing and fine-tuning multimodal/VLA models, including RL for embodied agents

• Proven ability to build scalable training and deployment pipelines for real-world robotic systems

• Expert programming skills in Python and C++, with PyTorch or JAX, focused on performance and rapid experimentation

• Hands-on experience with advanced physics simulators (Isaac, MuJoCo, Newton, etc.)

• Practical sim-to-real expertise, including system identification and robust domain transfer

• Direct experience with robotic hardware, multisensor systems, and manipulation tasks

• Ability to execute quickly, take ownership, and thrive in fast-paced environments

• Strong communication skills across research, engineering, hardware, and product teams

• Bonus strengths: knowledge of foundation models (flow/diffusion), differentiable simulators, top-tier publications, and open‑source contributions