Overview
About the role
Join the robotics and AI innovation lab of a major automotive manufacturer as a hands-on robot learning engineer focused on reinforcement learning. You'll use RL to make Vision-Language-Action (VLA) models and learned policies work better on real robots for manufacturing tasks: higher success rates, faster execution, and reliable recovery from failures. About 80% of the role is applied engineering on physical hardware and 20% is research and exploration. You'll report to the Staff Robot Learning Engineer. Mountain View, CA (on-site, 5 days a week) What you'll do
- Apply RL to improve manipulation success rate, cycle time, robustness and failure recovery on physical robots
- Fine-tune and evaluate VLA models and learned policies using approaches such as PPO, SAC, residual RL, policy adaptation and learned recovery policies
- Design reward functions and success detectors, including learned or VLM-based reward models
- Use sample-efficient methods for real-world training: RL from demonstrations, offline-to-online RL, human-in-the-loop corrections
- Set up safe real-robot RL: action limits, workspace constraints, automated resets, supervised exploration
- Diagnose and close sim-to-real gaps with domain randomization, system identification and real-world fine-tuning
- Run experiments directly on robots, including bimanual setups, and own evaluation, debugging and deployment
- Supervise data collection and catch data-quality problems in demonstrations and datasets
- Work with the simulation team on benchmarking and model evaluation
What you'll bring
- PhD in Computer Science, Robotics or a related field, or a Master's plus 3+ years of industry experience
- RL policies deployed or validated on physical robots. Simulation-only experience does not meet this requirement. Any robot counts: arms, legged, humanoid or mobile.
- Hands-on sim-to-real transfer experience, including diagnosing why simulated and real behavior differ
- Strong RL fundamentals: policy optimization, reward design, training and evaluation
- Python, PyTorch, Git and ROS/ROS 2
- Experience with NVIDIA Jetson or similar edge compute, and robotics MLOps tools (experiment tracking, dataset versioning, Foxglove or similar)
- Experience with sensor calibration, force-torque sensors or teleoperation data collection
- Comfort working hands-on with robots every day and owning experiments end to end
Nice to have
- Robotic manipulation experience, especially contact-rich or bimanual tasks
- Combining RL with pretrained VLA policies or imitation learning
- VLA models such as Pi0/Pi0.5, OpenVLA or GR00T
- Industrial or collaborative robots (Fanuc, Universal Robots) in manufacturing settings
|