Founding AI Research Engineer - Robot Learning

Origin • San Francisco, CA

3d • $100,000 - $300,000

About The Position

Origin is building Physical AI for the built world, starting with autonomous robots for interior construction. Its modular robots use Construction Action Models to learn and adapt to unstructured construction sites. The robots are deployed on live sites in New York City, supporting large-scale commercial projects by accelerating schedules and improving safety and predictability. Origin is backed by Tier-1 investors and aims to address labor shortages amid rising demand for housing, data centers, and manufacturing infrastructure.

The system uses a Multi Agent Action Expert architecture, combining classical precision algorithms with learned policies. This role expands the learned components while maintaining production safety and owns the full lifecycle of learned components on the OG-1 robot, from data collection and model training through edge deployment on Jetson AGX Orin. Every research project has a deployment milestone; this is not a lab position.

Requirements

• BS, MS, PhD in Computer Science, Robotics, Machine Learning, or equivalent experience shipping learned systems on physical robots.
• Strong Python and PyTorch skills, with comfort modifying research codebases and working directly with open-source VLA implementations.
• Experience in at least two of: imitation learning, reinforcement learning, vision-language models, robot learning from demonstration, or sim-to-real.
• Track record deploying machine learning on real hardware, including debugging policy failures on the actual robot.
• Working knowledge of ROS2 or equivalent robotics middleware.
• Experience working with simulation systems such as Isaac Sim.
• GPU profiling and optimization experience with TensorRT, ONNX, or CUDA; understanding of the impact of policy latency on contact control.

Nice To Haves

• Hands-on experience with VLA architectures such as π0/π0.5, OpenVLA, RT-2, or Octo, or foundation-model fine-tuning for robotics.
• Experience collecting teleoperation data and building DAgger/HG-DAgger pipelines.
• Experience with world-model architectures such as DreamerV3, V-JEPA, or latent dynamics models.
• Experience in construction, manufacturing, or contact-rich industrial domains.
• Publications at CoRL, RSS, ICRA, or NeurIPS; equivalent shipped work is also valued.

Responsibilities

• Train and deploy vision-language-action (VLA) models for contact-rich manipulation using the imitation learning infrastructure.
• Build data-collection pipelines, including teleoperation with GELLO, SpaceMouse, and VR; DAgger-style online correction; and demonstration curation.
• Research and prototype world models for surface-state prediction, spray dynamics, and anomaly detection.
• Design offline evaluation metrics that predict real-world finishing quality before deployment.
• Optimize models for edge deployment, including TensorRT compilation, latency profiling, and memory budgeting on dual Jetson AGX Orin systems.
• Design interfaces where learned policies propose actions and deterministic safety layers enforce constraints.
• Own the lifecycle of learned components, from data collection and training through deployment on the OG-1 robot.

What This Job Offers

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