Figure
Reinforcement Learning Engineer – Whole Body Control
San Jose, CAFrom $150kmidAdded 1 month ago
About this role
Figure AI is seeking a Reinforcement Learning Engineer to develop and deploy advanced RL algorithms for controlling their humanoid robots' whole body movements. You'll optimize sim-to-real performance, define evaluation metrics, and ensure robust control systems for general-purpose robotic tasks.
What you'll do
- Develop, train, and deploy reinforcement learning algorithms for whole body robot control
- Design optimal observation and action spaces and select appropriate model architectures
- Identify and resolve simulation-to-reality performance gaps
- Define, test, and evaluate performance metrics for learned policies
- Ensure robustness and reliability of the control stack
- Lead complex control projects and mentor junior engineers
What they're looking for
- Reinforcement learning algorithms (PPO, SAC, etc.)
- Dynamics and control theory, preferably legged robotics
- Hyperparameter tuning and cost function optimization
- Domain randomization, curriculum learning, reward shaping
- Behavior cloning and distillation techniques
- Project leadership and technical mentoring
- Python or similar programming languages
- Robotics simulation and control systems
Benefits
- Competitive salary: $200,000–$300,000 annually
- Full-time position with comprehensive total compensation package
- 5 days/week in-office collaboration in North San Jose, CA
- Work on cutting-edge humanoid robotics with human-level intelligence goals
Opens the application — the Jobs AI extension fills it for you. Set up autofill
Opens the official application on the employer’s site. No login required.
Figure
Figure develops advanced humanoid robots powered by AI technology. The company is hiring engineers across mechanical design, firmware development, manufacturing, quality assurance, and security to build and refine its autonomous robotic systems.
View all jobs at FigureLikely interview questions
- Walk us through a specific RL project where you tuned hyperparameters for PPO or SAC—what was your process for balancing exploration vs. exploitation, and how did you validate the results?
- Describe your experience with sim-to-real transfer in robotics. What domain randomization or other techniques have you used to close the gap, and what were the biggest surprises when deploying to real hardware?