Safety Validation Lead, Scenarios

Cssmerge; San Francisco (CA); On-site; USD 170,000 - 240,000; Full time

Job summary

Atoms is building the machines that power the next era of progress. We are seeking a senior safety validation engineer to own the safety scenario catalog and the evidence strategy for automated vehicle validation.

You will work with log-derived scenarios and NCAP-based protocols, build pipelines in Python, and shape the safety argument with rigorous, traceable reasoning. You will define execution modalities, manage regression suites, and ensure coverage across both real-world and simulated data,

Qualifications

• 6+ years in safety validation for automated vehicles or safety-critical domains.
• Owned scenario-based validation and catalog development.
• Hands-on with crash-avoidance protocols and adaptation.
• Strong data skills with large log data; Python; statistics.
• Familiarity with ISO 21448 (SOTIF), ISO 34502 and SAE J3016.
• Clear written communication.

Responsibilities

• Define criteria for safety-relevant scenarios and maintain taxonomy.
• Build a log-based scenario extraction pipeline from fleet data.
• Own crash-avoidance protocol suites and their test conditions.
• Define execution methods (SIL/HIL/on-road) and maintain regression suite.
• Translate triggering conditions into scenario-based evidence for safety case.

Skills

Safety validation

Python programming

Data analysis

OpenSCENARIO

OpenDRIVE

ISO 21448

SOTIF

Simulation

Technical writing

Vehicular safety

Tools

OpenSCENARIO

OpenDRIVE

Euro NCAP protocols

SAE J3016

Job description

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

• Scenario definition and taxonomy. Own the criteria that make a scenario safety‑relevant and the criticality measures that decide it. Maintain the taxonomy and the abstraction levels — functional, logical, concrete — and keep the catalog coherent as it grows rather than letting it accrete.
• Log‑based scenario extraction. Build the pipeline that mines fleet data for safety‑relevant events — disengagements, interventions, near‑misses, anomalous interactions — and converts them into parameterized, reusable scenarios. Own the criticality metrics used to surface candidates, such as time‑to‑collision, post‑encroachment time, and margin to the drivable envelope, and be clear‑eyed about what those metrics miss.
• Crash‑avoidance protocol suites. Own the implementation and upkeep of standardized suites derived from the crash‑avoidance portions of NCAP and equivalent published protocols — AEB car‑to‑car, AEB vulnerable road user (pedestrian, cyclist, motorcyclist), lane support, and emergency steering cases — including test conditions and pass/fail criteria.
• Execution and evidence. Define how scenarios run — resimulation, SIL and HIL, vehicle‑in‑the‑loop, closed course, on‑road — and what evidence each method can and cannot produce. Own the regression suite that keeps known failures from returning.
• Interfaces. Take triggering conditions and hazards from functional safety and field events from safety operations, and return scenario‑based evidence they can rely on. Your output is a substantial part of the validation argument in the safety case.

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 AI 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. Concretely, we want someone who:

• Uses LLM‑based and purpose‑built tooling 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, and drafting and maintaining validation documentation.
• Builds their own tooling rather than filing tickets for it. Writes the extraction pipelines, the coverage analysis, and the reporting, and keeps them running.
• Understands where AI assistance is legitimate and where it is not. A generated scenario set is a hypothesis to verify, never coverage. Validation claims require human judgment and traceable justification, and you should be rigorous about that boundary while still capturing the leverage.
• Can reason about the limits of the evidence — simulation fidelity, distribution shift between logged and simulated behavior, and what a passing suite does and does not entitle you to claim.

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

• 6+ years in safety validation, verification, or test engineering for automated vehicles, ADAS, or another safety‑critical domain, with deep hands‑on experience rather than test management alone.
• Demonstrated ownership of scenario‑based validation on a real program: you have built or substantially shaped a scenario catalog and defended its sufficiency to someone who pushed back.
• Hands‑on experience 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.
• Strong data ability. You work directly with large volumes of vehicle log data: querying it, building extraction and clustering pipelines, and forming a defensible position from it. Strong Python; comfortable with the statistics of exposure, rare events, and coverage.
• ASAM OpenSCENARIO and OpenDRIVE, hands‑on — you have authored and debugged scenarios in them. ISO 21448 (SOTIF) as the frame for why safety‑relevant scenarios matter, and ISO 34502 as the scenario‑based safety evaluation framework. Familiarity with SAE J3016.
• Simulation experience, and specifically an informed view of where simulation evidence is credible and where it is not.
• Clear written communication. Coverage arguments live or die on whether a reader can follow the reasoning.

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