Robotics Product Engineer: Full-Stack, ROS 2 & Real-Robot Deployment | San Francisco (On-site) | Up to $275K + Equity

"I'm sorry, Dave. I'm afraid I can't do that."HAL 9000, 2001: A Space Odyssey

HAL had one job and declined it. Decades later, most robots are far more polite about failing. They simply stop, stall, or drive gently into a table leg while a very capable model waits somewhere in the cloud. Nobody is being menacing. Something between the research and the robot just isn't working.

This role exists to close that gap.

The short version

Our client is a San Francisco AI company building the product that lets robotics teams run the newest models on their own machines. They need a hands-on product engineer who has done this in the physical world: someone who has taken a model or a policy off a researcher's laptop, put it on a real robot, watched it fail in an interesting way, and then fixed it.

You'll be an individual contributor on a small, flat team reporting directly to the CTO. The work is in person in San Francisco, with actual robots in the office to test on. Base salary is $225K to $275K, plus equity.

Why this job is different

Most robotics software roles put you on one side of a wall. You either sit near the hardware and never meet a customer, or you sit near the customer and never touch the hardware. This one removes the wall.

Half of the work faces outward. You'll spend real time with the robotics teams who use the product, sitting with them while they work, seeing where they get stuck, and working out what would make next week better than this one. Their failures are your raw material.

The other half faces inward. You take what you saw and decide what it deserves: a quick patch tonight, a proper feature next sprint, or a new capability in the platform that every customer will benefit from. Then you build it yourself, across the interface, the backend, the APIs and the infrastructure underneath.

Few engineers get to own the full loop from "the arm keeps dropping the cup" to "shipped, and it doesn't any more". Here, that loop is the job.

Coit Tower after dark. San Francisco is where this one happens, in person.

An illustrative week

On Monday you're in a customer's lab, watching a manipulation policy that scored beautifully in evaluation hesitate over a real object. By Tuesday you've traced it, and the model was never the problem: it was latency between the robot and the inference endpoint. On Wednesday you ship a fix and see that three other customers will hit the same thing, so by Thursday it has become a platform capability with a proper interface. On Friday a researcher hands you something new and asks whether it can run on the arm by the window. You say yes, mostly because you want to find out.

What you'll do

Sit with robotics customers, study what's breaking, and turn those observations into product decisions.Work side by side with the research team so that new results run on physical robots, not only in a paper or a simulator.Make the call on scope (patch, feature or platform), then ship it front to back.Chase problems wherever they hide: robot software, networking, cloud, model inference. The bug rarely lives where the symptom shows up.Build simulation and visual tooling that lets customers set up environments, inspect what a robot actually did, and understand why.Test your own work on the robots in the SF office. If it only works in simulation, it doesn't work yet.

What they're looking for

This is the scorecard, stated plainly:

You've shipped production software end to end. From first idea to real people using it. Not a slice of a system; the whole arc.You have recent, hands-on robotics experience. A couple of years is plenty, but it has to be real and it has to be recent. Models or policies running on physical machines is the signal.You build across the stack. Front end, backend, infrastructure. You don't wait for another team to unblock you.You've worked directly with users to decide what to build. You're comfortable in the room with a customer, and you come back with a sharper product rather than a list of requests.You've turned research into working software. You've partnered with researchers and you know the distance between a promising result and a dependable product.

The Ferry Building and downtown at night.

What will make you stand out

You've deployed learned models or policies on real robots.You know your way around ROS 2, MuJoCo or Isaac Sim.You've built interactive 3D, browser graphics or simulation interfaces.You've worked on policy serving, teleoperation or robot evaluation workflows.You've spent time at a fast-growing model inference or AI infrastructure platform. Engineers from that world tend to be an excellent match.You've seen both a large company and a startup, and you know what each is good for.

Who thrives here

People who like getting their hands on hardware. People who find a flaky network link just as interesting as a model architecture. People with low ego who'll happily pick up the unglamorous task, because it's the one standing between a customer and a working robot.

A word on fit

Two things matter enough to say clearly.

First, adjacent experience isn't the same thing. Strong work in computer vision, motion planning or simulation is valuable, but the team is specifically looking for people who have worked with real robots, in the real world, recently.

Second, this is a builder's role. There is no team to manage and no plan to add one beneath you. If your next step is people leadership, this isn't it. If your next step is the most interesting engineering problem you've had in years, read on.

The process

Sensible, practical and quick:

An intro conversation with the Head of Talent.A 20 to 30 minute chat with an engineer about products you've built.A 45 minute technical session. Practical and in-domain, with no LeetCode at any stage.An onsite in San Francisco to meet the team. If you're not local, they'll fly you in and cover your night's stay.

The Transamerica Pyramid from Chinatown.

The details

Location: San Francisco, CA. In person, with the team and the robots.Compensation: $225K to $275K base, plus equity.Relocation: Open to candidates willing to move to San Francisco.Reporting line: Directly to the CTO.Role type: Individual contributor.

Why now

Robot models are improving faster than the tooling around them. The teams building robots have capable models and capable hardware; what they're missing is the product in between, the layer that makes a new model something you can run, inspect and trust on a machine that moves. The engineer who builds that layer will shape how a whole industry gets its robots to do what they're asked.

HAL couldn't do that. You can.

Apply using the form on this page.