Research Engineer, Robot Learning

1X • San Carlos, CA

1d

About The Position

Develop sim-to-real reinforcement learning (RL) training pipelines for humanoid robots. Train and test robot policies using reinforcement learning for in-home humanoid robots, and optimize training pipelines to improve performance and training speed. Model physical systems—including robot dynamics, sensor behavior, and contact parameters—to reduce the gap between simulation and physical systems. Collect, clean, and retarget human motion data for use in RL training pipelines. Support testing and debugging of RL policies in simulation and on physical hardware.

Requirements

• Master’s degree or foreign degree equivalent in Robotics, AI/Machine Learning, Mechanical Engineering, Electrical Engineering, Computer Science, or a related field.
• One year of experience in a Robotics and Machine Learning occupation or a related occupation in the job offered.
• Experience or education in rigid-body dynamics and/or dynamical systems for training, deploying, and testing reinforcement-learned policies for humanoid robots.
• Ability to implement production-grade Python code for reinforcement-learning and other machine-learning training pipelines.
• Knowledge of reinforcement-learning algorithms, techniques, and sim-to-real approaches for robotics and human-like capabilities, including whole-body loco-manipulation.
• Ability to implement and debug real-time C++ software deployed on robots subject to real-time constraints.
• Ability to model physical systems, including multi-body hybrid dynamic systems with contact, in simulation; identify simulation-to-hardware gaps; and understand how to accurately model hardware in simulation.
• Root-cause analysis skills for robotic systems and ability to troubleshoot issues quickly.
• Knowledge of forward and inverse kinematics for humanoid-system training using human motion data.

Responsibilities

• Develop sim-to-real reinforcement learning training pipelines for humanoid robots.
• Train and test reinforcement-learning robot policies for in-home humanoid robots.
• Optimize training pipelines for improved performance and training speed.
• Model physical systems, including robot dynamics, sensors, and contact parameters, to reduce simulation-to-hardware gaps.
• Collect, clean, and retarget human motion data for reinforcement-learning training pipelines.
• Support testing and debugging of reinforcement-learning policies in simulation and on physical hardware.

What This Job Offers

• Job Type: Full-time
• Career Level: Mid Level