Safety Validation Lead, Scenarios

About the role

Who we are Atoms is building the machines that power the next era of progress. Over the last decade, software has transformed the digital world. But the physical world, where food is made, minerals are mined, goods are moved, and industries are run, remains far less intelligent, far less efficient, and far more constrained. We’re changing that. Atoms builds Physical AI— real-world robots for the industries that move civilization forward, starting with food, mining, and transport. Our systems are designed to understand, predict, and control the real world with precision, turning complex physical operations into something more reliable, more scalable, and more productive. This work requires more than robotics. It requires deep integration across hardware, software, AI, operations, manufacturing, and real estate. We don’t just build machines in a lab. We deploy them into real environments, operate them, learn from them, and improve them until they work at scale. We are roboticists, engineers, operators, and builders. We believe the next great technology companies will not only transform information, but the physical systems that shape everyday life. If you want to work on hard problems with real-world impact, join us. About the role This role owns the safety-relevant scenarios we validate against. You will define what makes a scenario safety-relevant, build the catalog, and own the argument that it is sufficient. The catalog draws from two sources that behave very differently. Log-based scenarios come from real events, real interactions, and real distributions, and they bring the problem of finding the few that matter inside a very large volume of ordinary driving. NCAP-based scenarios come from the crash-avoidance portions of published consumer test protocols: precisely specified, externally credible, and designed for a driver-assist framing that has to be adapted before it means anything for an automated platform. You will own both, and you will own the harder question of how they fit together into one coverage argument. This is a technical individual contributor role. You will build the pipelines and write the analysis yourself, and the coverage position you take becomes part of the safety case. What you’ll do - Relevance and coverage. Define what makes a scenario safety-relevant for a given slice of the operating domain, and own the coverage position that follows. Keep the taxonomy and the abstraction levels coherent as the catalog grows rather than letting it accrete. - Tooling. Build the tooling that turns relevance and coverage into a repeatable query rather than a judgment reconstructed from scratch each time, and that shows where a catalog has gone stale. Other engineers run it; you build it. - Both sources. Own log-based extraction from fleet data and the criticality measures that surface candidates from it, and own the standardized crash-avoidance suites and the adaptation they need before they mean anything here. - Execution. Define how scenarios run and what a result from each method can and cannot support, and own the regression suite that keeps known failures from returning. - Interfaces. Designing the scenario is where the argument gets set up: it fixes which claim is being exercised, under what conditions, and what a result would mean. Whether a given run passes is evidence, and it is produced elsewhere — your part is that the scenario was the right one to run and that its outcome is interpretable against the claim. Working AI-native We expect this role to be materially more productive than the same role was three years ago, and we expect AI tooling to be the reason. We are adopting systems purpose-built to accelerate safety analysis — scenario mining and categorization, criticality assessment, parameter space search, and documentation — and this role is expected to put them to work and shape what they become. - Use them as a working instrument in the analysis itself: classifying and clustering large volumes of log events, proposing and stress-testing scenario variants, cross-checking a catalog for gaps, mapping published protocols onto our own taxonomy. - Build your own tooling rather than filing tickets for it, and keep it running once other people depend on it. - Be rigorous about the boundary. A generated scenario set is a hypothesis to verify, never coverage. Validation claims require human judgment and traceable justification. We would rather hire a strong validation engineer who is curious and moving fast on AI tooling than someone who has the vocabulary but not the practice. Be prepared to show us how you actually work. What we’re looking for - 8+ years in safety validation, verification, or test engineering for automated vehicles, ADAS, or another safety-critical domain, hands-on rather than test management alone. - You have built or substantially shaped a scenario catalog scoped to a defined operating domain, and defended its sufficiency to someone who pushed back. - Hands-on with the crash-avoidance side of published consumer test protocols — Euro NCAP's Crash Avoidance protocols in the 2026 series, the ADAS component of US NCAP, or equivalent — and the judgment to adapt rather than transplant them. - ASAM OpenSCENARIO and OpenDRIVE, hands-on; ISO 21448 (SOTIF) and ISO 34502; and an informed view of where simulation evidence is credible and where it is not. - Strong Python and direct work with large volumes of vehicle log data, plus tooling other engineers ran for their own work rather than only your own. - Fluency with claim, argument and evidence as distinct things. You can say what a result actually supports, and you do not present evidence as though it were the claim. Nice to have - Proving ground or closed-course test execution experience, including instrumented targets and test equipment. - Experience with falsification, adversarial scenario search, or criticality-driven parameter sampling. - Experience contributing scenario-based evidence to a safety case defended to an external party. - Background in traffic safety research, crash data analysis, or naturalistic driving studies. - Experience building validation infrastructure — scenario databases, resimulation at scale, CI for autonomy. Why join us At Atoms, you’ll work on one of the defining challenges of our time—bringing automation into the physical world to drive real, lasting impact. We exist to uncover valuable unknown truths and turn them into progress, which means constantly pushing beyond what’s known and building what doesn’t yet exist. The work is ambitious and often challenging, but it’s grounded in a shared sense of purpose and a team committed to seeing it through together. Our work only matters if it serves others, and we know that meaningful progress depends on the trust of the people we serve and the strength of our team—so we invest in both, creating an environment where you can do your best work and grow. What else you need to know This role is based in our San Francisco office. Atoms is a company driven by invention and continuous change - we are constantly reimagining our industries, building new products, and refining how we operate. We do our best work together. That’s why all of our office-based teams work onsite, five days a week. The base salary range for this role is $182,000 - $231,000 per year. Actual compensation will be determined on an individual basis and may vary depending on experience, skills, and qualifications. Base salary is just one part of your total rewards package. You may also be eligible for equity awards. Benefits Summary (USA Full-Time Exempt Employees): - Medical, Dental, Vision, Disability, and Life Insurance - Flexible Spending Account / Health Savings Account Options - 401(k) - Equity - Sick Time, Unlimited Flexible Time Off, and Paid Holidays - Paid Parental Leave - Pre-Tax Commuter Benefit Plan - Team lunch in our SoMa office every Tuesday and Thursday Benefits are subject to change at the company's discretion. Atoms accepts applications on an ongoing basis. Ready to join us as we serve those who serve others? #LI-Onsite