Robot Test Engineer

Department: Engineering

Employment Type: Full Time

Location: San Francisco

Description

Reactor is building the inference platform for real-time world models. Our goal is to democratize access to real-time intelligence to unlock next-generation use cases in media, physical AI and omni-directional models. We already run the largest and fastest catalog of world models on the market, and work closely with all major model labs to run and optimize their world models, including Alibaba, Amazon, and NVIDIA. Our team is made of alumni from leading firms such as Apple, Meta, Amazon, Microsoft, Luma AI, Adobe, Stability AI, and more. We've raised $74M led by Lightspeed, with Amplify Partners, WNDR, NVIDIA, Sapphire, and other Tier 1 investors participating. Jeffrey Katzenberg, co-founder of DreamWorks Animation, sits on our board of directors.

Robot learning moves fast, and you’ll be the one who proves a new model actually works on a real robot.

You'll help build out our robot testing fleet, get the newest policies (vision-language-action models, imitation learning checkpoints) running on real arms, and ship the live demos that show what our software stack can do. When a trial fails, you'll work out why: a loose camera mount, a flipped coordinate frame, a timing bug in the control loop, or a model that isn't as good as its benchmark says.

You'll work directly with our researchers and engineers, and you'll help decide what we test and how. As the lab grows, you'll build the tools and procedures that make testing faster, safer, and repeatable.

We're in person in San Francisco.

What You'll Do

• Help build out our robot testing fleet as it grows: new arms, new setups, new test cells.
• Build and run live demos of our software stack on robots, for partners and launches.
• Deploy learned policies on robot arms and other hardware. Wire up cameras, configure inference, and confirm the model gets the right observations and sends the right commands.
• Design manipulation tasks and run repeated trials. Compare policies and checkpoints on task success, interventions, and failure modes, so we know whether a new checkpoint actually grasps better or just got lucky.
• Stress-test behavior across objects, lighting, placements, and starting conditions.
• Debug failures end to end, across hardware, calibration, policy inputs and outputs, latency, and control software.
• Build and maintain robot setups: grippers, camera mounts, and custom fixtures when an experiment calls for them.
• Capture videos, logs, and configs so anyone on the team can reproduce a run, then help test the fix.
• Write Python tooling to launch experiments, replay and inspect trajectories, and compare results.
• Keep the robots ready to go, and set the safety procedures for trying new policies on hardware.

What We're Looking For

You've built, run, or repaired robots and liked it enough to want to do it every day. That could be from a job, a research lab, a robotics competition, or your own garage.

You're comfortable with:

• Python, Linux, and Git, and getting unfamiliar software running without a guide.
• Physical hardware, and tracking down mechanical, electrical, and software problems.
• Isolating variables and testing hypotheses when a robot does something unexpected.
• Running careful, repeated experiments and knowing exactly what changed between runs.
• Explaining what you found, what you ruled out, and what you still don't understand.

Most of all, you're curious about robot learning, patient with hardware, and you push through problems on your own. You enjoy both the first time a new model moves an arm and the follow-up work of finding out how reliably it does it.

Bonus Points

• Running VLAs or learned policies on real robots (π0, OpenVLA, GR00T, ACT, Diffusion Policy, or similar).
• LeRobot, teleoperation, demonstration collection, or policy evaluation.
• Camera calibration, coordinate frames and transforms, inverse kinematics, and robot control.
• ROS / ROS 2, PyTorch, or model inference tooling.
• CAD, 3D printing, basic electronics, or building custom test fixtures.

How to Apply

Show us something you've built or debugged. Project videos, GitHub repos, and photos of your setup all count. Tell us what went wrong and how you figured it out.