Research Engineer, Robotic Learning

1X • San Carlos, CA

3d • $166,566 - $350,000

About The Position

Design simulation environments for NEO (name of robot) and teach NEO (robot) to learn new capabilities via reinforcement learning (RL) algorithms. This enables our robots to be safe and robust in a variety of conditions. Full-stack engineering, from data engineering to model architecture design to shipping polished products. Train NEO to do a diverse set of manipulation and locomotion tasks. Close the sim2real gap between policies trained in simulation and real. Work with controls, Quality Assurance (QA,) and data collection teams to ship reinforcement learning (RL) policies to the production fleet. Deploy skills trained with reinforcement learning (RL) into home environments.

Requirements

• Bachelor’s degree or foreign degree equivalent in Computer Science, Mathematics, Software Engineering, Robotics, or related field
• Four (4) years of experience in Software Engineering or related role
• Experience and/or education must include training large-scale imitation-learning models for motion planning
• Experience and/or education must include curating expert demonstration datasets, designing loss functions, and deploying policies in real-world or simulated environments
• Experience and/or education must include implementing DAgger (Dataset Aggregation), including iteratively collecting additional expert demonstrations in response to model errors, to reduce covariate shift and improve policy robustness
• Experience and/or education must include generative architectures such as diffusion models, variational autoencoder, or applied to planning
• Experience and/or education must include model architecture selection and stable training on high-dimensional data
• Experience and/or education must include architecting and executing distributed neural-network training across large GPU clusters
• Experience and/or education must include data/model parallelism, gradient synchronization such as NCCL, fault tolerance, and efficient I/O for multi-TB datasets
• Experience and/or education must include building and optimizing low-latency, real-time software systems
• Experience and/or education must include use of C++ for hard-real-time applications, with hands-on knowledge of multithreading, lock-free data structures, and real-time OS constraints

Responsibilities

• Design simulation environments for NEO
• Teach NEO to learn new capabilities via reinforcement learning (RL) algorithms
• Train NEO to do a diverse set of manipulation and locomotion tasks
• Close the sim2real gap between policies trained in simulation and real
• Work with controls, Quality Assurance (QA,) and data collection teams to ship reinforcement learning (RL) policies to the production fleet
• Deploy skills trained with reinforcement learning (RL) into home environments

What This Job Offers

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