The Role
As a member of technical staff, you will bridge the World Model's predictions with physical robot capability, engineering and training policies that solve physical tasks. You are an expert at the intersection of robotics and reinforcement learning, focused on turning the World Model's understanding of physical dynamics into policies that perform reliably on deployed hardware. You own the complete policy development cycle: data collection, training, evaluation, and iteration on hardware.
Job Responsibilities
- Train and deploy policies that solve physical tasks on robotic hardware, building on the World Model's predictions and collected sensor data.
- Design and instrument data collection on robotic platforms, including task design, sensor synchronization, and capture procedures robust to unstructured physical environments.
- Build evaluation frameworks that measure task success, physical accuracy, and generalization of trained policies on hardware.
- Close the loop between simulation and deployment, feeding results from hardware evaluation back into the pretraining, post-training, and control research tracks.
Minimum Qualifications
- Direct, hands-on experience training and deploying policies (RL, imitation learning, or similar) on physical robotic systems.
- Experience working with robotic hardware and sensor data.
- Expert-level understanding of deep learning frameworks (e.g., PyTorch, JAX); Python required, Rust/C++ a plus.
Preferred Qualifications
- PhD in Robotics, Computer Science, Machine Learning, or a related field.
- Track record of shipping policies or control systems for manipulation or navigation tasks.
- Familiarity with ROS/ROS2 or similar robotics middleware.