About the job
The AI Controls team builds high-rate learned controllers that let Digit move robustly, efficiently, and safely in dynamic environments. As an AI Controls Engineer, you’ll develop and deploy reinforcement learning policies across humanoid locomotion, whole-body control, and manipulation—integrating perception to enable collision-free, perceptive motion in the real world.
Responsibilities
Design, train, and deploy robust RL policies for locomotion, manipulation, whole body control, and dynamic interactions with the environment.
Integrate perception into RL policies to achieve obstacle-aware, collision-free motion, and perceptive manipulation.
Develop and maintain core RL infrastructure, including scalable training pipelines and evaluation frameworks.
Design and implement new simulation environments and tasks to support training and evaluation of control policies.
Collaborate with on-robot software and deployment teams to ship production-quality policies to Digit.
Qualifications
Minimum
4+ years of experience developing and deploying RL policies for robotics applications.
Strong Python skills and hands-on experience with a deep learning framework such as PyTorch.
Experience designing reward functions, tuning hyperparameters, and implementing exploration strategies to solve complex control tasks.
Experience with perception-in-the-loop control, integrating real-time sensory inputs for reactive or adaptive behaviors.
Proven experience deploying reinforcement learning policies on real-world bipedal or quadrupedal robots.
Familiarity with robot simulation environments (e.g. Mujoco-Warp, Isaac) and sim-to-real transfer.
A collaborative approach and the ability to deliver safe, high-quality software in a fast-paced environment.
Preferred
Advanced degree (MS or PhD) in Robotics, Computer Science, or a related field.
Experience with contact-rich manipulation, including force-torque or tactile sensing.
Familiarity with policy distillation (e.g. teacher–student) for transferring state-based policies to perception-driven ones.
Publications in top ML or robotics conferences (e.g. NeurIPS, ICML, CoRL, RSS, ICRA).