About Origin Origin is building Physical AI for the built world - starting with autonomous robots for Interior Construction. We are building Construction Action Models which allows our modular robots to learn, adapt and work in unstructured construction job site. Our robots are already deployed on live sites in New York City, helping accelerate schedules for large-scale commercial projects while improving safety and predictability on the job site. Backed by Tier-1 investors, Origin is working to close the gap between America’s surging demand for housing, data centers, and manufacturing infrastructure, and the construction industry’s growing labor shortage. The Role Our system runs a Multi Agent Action Expert architecture: classical precision algorithms orchestrated alongside learned policies. The job is systematically expanding the learned components while keeping the system production-safe. You own the full lifecycle of learned components on OG-1: from data collection and model training through edge deployment on Jetson AGX Orin. Every research project will have a deployment milestone. This is not a lab position. What you will own? • Define the technical roadmap for Robot Learning and Embodied AI. • Build and deploy learned policies for real-world mobile manipulation and contact-rich tasks. • Develop imitation learning, reinforcement learning, VLA, and learning-from-demonstration systems. • Fine-tune and adapt open-source VLA/foundation models for our robot platform. • Build scalable teleoperation → dataset → training → evaluation → deployment loops. • Develop DAgger / HG-DAgger and human-in-the-loop data collection pipelines. • Build simulation environments and training pipelines using NVIDIA Isaac Sim / Isaac Lab. • Develop sim-to-real strategies including domain randomization, system identification, and real-world policy adaptation. • Explore world models and latent dynamics models for planning, prediction, and policy learning. • Integrate learned policies with our existing ROS2 perception, planning, manipulation, control, and safety stack. • Optimize inference for deployment on edge GPUs using TensorRT, ONNX, CUDA, profiling, quantization, and related techniques. • Debug policies on physical robots: latency, observation drift, calibration errors, distribution shift, contact instability, action representation, control frequency, and hardware-induced failures. • Establish rigorous evaluation for learned systems across simulation, replay datasets, and physical robot experiments. • Build and mentor the Robot Learning team as we scale. Our broader robot architecture already spans perception, manipulation, planning, ROS integration, simulation, force/impedance control, quality assessment, and autonomous recovery. Requirements • BS/MS/PhD in CS, Robotics, ML, or related field from Stanford, MIT, UC Berkeley, CMU, Georgia Tech, ETH Zürich, or UPenn, or equivalent exceptional experience shipping learned systems on physical robots. • PhD: minimum 2 years relevant experience. Without PhD: minimum 5 years relevant experience. • Strong Python and PyTorch; comfortable modifying research codebases and open-source VLA implementations. • Experience in at least two of: imitation learning, RL, VLA/VLMs, robot learning from demonstration, sim-to-real. • Track record deploying ML on real robots — not just training policies, but debugging why they fail on actual hardware. • Working knowledge of ROS2 or equivalent robotics middleware. • Experience with simulation systems such as NVIDIA Isaac Sim / Isaac Lab. • GPU inference profiling and optimization (TensorRT, ONNX, CUDA); understand the impact of policy latency on real-time robot control. Strong Plus • Hands-on with VLA architectures such as π0/π0.5, OpenVLA, RT-2, Octo, or robotics foundation-model fine-tuning. • Teleoperation data collection and DAgger / HG-DAgger pipelines. • World models such as DreamerV3, V-JEPA, or latent dynamics models. • Experience with contact-rich manipulation, construction, manufacturing, or industrial robotics. • Publications at CoRL, RSS, ICRA, NeurIPS — valued, but equivalent shipped work on real robots counts.