Job Description
About the Organization
Join the team building GM's next generation of autonomous driving, starting with eyes-off driving on the highway and growing toward point-to-point autonomy. The goal is autonomous capability that is safe, reliable, comfortable, and scalable across GM's vehicle portfolio. Autonomy Evaluation helps the AV program figure out where the driver still falls short. Every mile our vehicles drive, on the road or in simulation, carries that signal. We turn it into answers: what is limiting the system, how big each problem is, and what fixing it is worth. Those answers steer what GM's autonomy teams build next and inform software release decisions. We partner closely with Autonomy, Simulation, Systems, Safety, and Software Validation.
The Role
As a Principal Engineer in Autonomy Evaluation, you will own how we evaluate the autonomous driver, from overall driving behavior down to individual subsystems. You'll take on questions like: How do we measure whether the car drives well, not just safely? When a behavior goes wrong, which subsystem caused it, and how much does that error matter on the road? How do we find the failures we don't know to look for? You'll turn the answers into methods other teams adopt, priorities other organizations commit to, and evidence that release decisions rely on. You'll prototype and land code yourself, and you'll grow the Staff and Senior engineers who carry the work forward. You are a senior individual contributor with no direct reports.
What You'll Do
Set the multi-year technical direction for evaluation. Decide which questions matter most, which methods to invest in, and which to retire.
Invent new ways to judge overall driving behavior, such as learned or VLM-based judges of driving quality, comparison with how expert humans drive, counterfactual and closed-loop analysis, and discovery of unknown failure modes in the long tail.
Define how we evaluate subsystems such as perception, prediction, planning, and localization, including learned components, with metrics that reflect each subsystem's real effect on driving rather than accuracy alone.
Trace bad driving back to the subsystem or interaction that caused it, and measure how subsystem errors add up to driving outcomes.
Define how we tell whether a software or model change actually improves driving, including regression detection and version-to-version comparison, so engineering teams and release reviews get trustworthy answers quickly.
Turn observed behavior into measured, ranked problems. Size issues with metrics and scene classifiers, build a Pareto of the problem space, and drive fixes with the teams that own them.
Own the scientific rigor of evaluation: sampling, statistical confidence, metric validity, and agreement between simulation and road results. Make every result's assumptions and limits explicit, and resolve the hardest disagreements between signals, such as simulation versus road or a subsystem metric versus a driving metric.
Turn results into clear recommendations for engineering and leadership: what changed, how confident we are, and what to do next.
Mentor Staff and Senior engineers through research and design reviews and written technical direction.
Your Skills & Abilities
10+ years of experience in autonomous systems, robotics, ML evaluation, or applied data science, with a track record of turning analysis into decisions and shipped improvements.
A track record of creating new evaluation or measurement methods that were adopted beyond your own team.
Deep experience designing statistically sound evaluations of dynamic systems, including sampling, confidence intervals, experimental design, hypothesis testing, and the precision and recall of classifiers on time-series data.
Strong production Python and SQL skills, including testing and code review.
A record of starting from raw observations, defining and sizing the problem, and driving it to resolution with data that others trust.
Proven ability to…