About the Role
You will own reinforcement learning across our stack, from locomotion policies to RL applied on top of learned behaviors. The research problems here are open ones: whole-body control under contact, sim-to-real transfer that holds up in environments we have never seen, and learning from a growing body of real-world robot experience.
This is a role for someone who wants their research to run on hardware in the field rather than on a benchmark, and who is early enough in a team's life to set its technical direction.
What You'll Do
Own RL across our stack, from locomotion to learned behaviors, and set the technical direction for how we apply it.
Track where the field is going, and make calls about which directions are worth betting on and which are noise.
Advance the state of the art in RL for embodied systems, and ship the results on real robots.
Apply RL to improve learned behaviors beyond what demonstrations alone can reach, including fine-tuning imitation-learned and pretrained policies.
Work on whole-body control, where the robot maintains balance while applying and reacting to external forces.
Learn from logged fleet data, applying offline and hybrid RL to experience collected from robots in the field.
Own reward design, curriculum, and exploration, and the diagnosis of why a policy is failing.
Drive sim-to-real transfer, working with Simulation on fidelity and randomization strategy.
What We're Looking For
PhD in reinforcement learning, robotics, or a closely related field from a group doing leading work, or equivalent research experience at a frontier lab.
Depth in RL fundamentals: policy optimization, value methods, model-based and offline RL, and the practical failure modes of each.
Research contributions that others build on: publications at leading venues, or widely used open source work. You have moved the state of the art, not only kept up with it.
Experience training policies that ran on physical systems, and the judgment that comes from watching them fail.
Strong engineering skills. RL at scale is an infrastructure problem as much as an algorithms problem.
Experience with partial observability and environment variability, where the policy has to work in conditions it was not trained on.
Bonus Points
Legged locomotion, particularly humanoid or bipedal.
Whole-body control, contact-rich tasks, or force-aware policies.
RL fine-tuning on top of imitation-learned or pretrained policies.
Experience at a leading AI or robotics lab.