Reinforcement Learning Engineer, Whole Body Controls, Optimus

Job description

Job responsibilities

• Develop end-to-end reinforcement-learning policies for whole-body control spanning locomotion and manipulation: walking, balancing, disturbance recovery, and manipulating objects
• Design reward functions, action and observation spaces, and curricula; tune and scale reinforcement-learning training
• Improve the sample efficiency, exploration, stability, and sim2real robustness of learned policies
• Build large-scale, simulation-based training pipelines
• Evaluate policies both in simulation and on hardware, and ship them to a fleet of bots

Job requirements

• Proficiency in Python (numpy, pytorch)
• Strong reinforcement-learning fundamentals, including policy-gradient and actor-critic methods, with hands-on experience training reinforcement-learning agents
• Curiosity about physics and robotics and an eagerness to learn control and dynamics on the job; no prior robotics experience required
• Preferred: large-scale or distributed reinforcement learning and massively parallel simulation (e.g. mjlab, IsaacLab); reward shaping, curriculum learning, or sim2real; any robotics or control exposure
• Equivalent experience through projects, research, or coursework is acceptable