Embodied AI Engineer, VLA Deployment

Matter is building the AI‑native autonomy stack for physical manufacturing. We own and operate factories as controlled learning environments—collecting data from every stage of production to train and validate AI systems that run on real hardware.

We are not building robots for a demo. We are building autonomous production systems for real customer programs in robotics, defense hardware, and consumer electronics.

We are hiring an Embodied AI Engineer to own the deployment and production hardening of VLA and learned robotic policies on our factory floor. This role bridges the gap between research‑grade model performance and the reliability requirements of a production manufacturing environment. You will take models from the training pipeline, validate them in simulation, and get them running on physical robots handling real parts.

The job is to make the models work—not in the best case, but in the median case, at tolerance, at the end of a 10‑hour shift, with part variation and lighting changes and calibration drift.

• Deploy and product ionize VLA models (e.g., OpenVLA, π0, Octo, or internal variants) on physical robot arms in assembly and test workcells.
• Own the Sim2
  
  Real validation workflow: verify that models trained in NVIDIA Isaac Sim or Mu Jo Co  transfer reliably to physical hardware before production deployment.
• Develop and tune the action prediction layer: end‑effector pose generation, trajectory planning, grasp synthesis, and force‑torque feedback integration.
• Fuse learned policies with traditional control methods to meet manufacturing safety and repeatability requirements (ISO 10218).
• Build robustness protocols for production edge cases: part variation, lighting drift, sensor calibration, wear‑induced performance degradation.
• Instrument deployed robots for continuous learning: collect failure cases, annotate edge cases, and feed them back into the training pipeline.
• Collaborate with the Safety Shield layer, ensuring probabilistic AI outputs are bounded by deterministic safety constraints.

• Proven track record deploying robot learning systems on physical hardware in real environments (not just lab or simulation).
• Deep familiarity with VLA model architectures and the operational gap between benchmark performance and production reliability.
• Experience with ROS2‑based systems, manipulation planning, and robot control stacks.
• Strong understanding of perception systems: camera calibration, point cloud processing, force‑torque sensing.
• Ability to debug across the full stack: model outputs, controller behavior, sensor noise, hardware faults, and communication latency.
• High tolerance for ambiguity; you thrive in environments where the answer is “go figure it out on the factory floor.”

• PhD or graduate experiences in robotics, machine learning, or a related field.
• Experience with safety‑critical robotic systems and functional safety standards.
• Background in manufacturing, industrial automation, or precision assembly.
• Familiarity with MARL or multi‑robot coordination for factory‑scale autonomy.

Most embodied AI roles give you a lab, a tabletop, and a list of benchmarks. At Matter, you get a 54,000‑sq‑ft factory with live production programs, modular automation built for 100% data collection, and a Sim2

Real pipeline connected to real customer hardware. The models you deploy are not demos. They run on production programs for companies building the next generation of robots and defense systems.

If you want to know what it takes to get VLA models from paper to production, this is where you find out.