Staff Probabilistic Risk Assessment Engineer

• We are hiring for a Senior Autonomy Safety Engineer to join the team. This will be used to support our safety claims about safe perception, motion planning, and control for autonomous robots deployed in both commercial freight and defense logistics environments
• Kodiak’s Autonomy Safety team owns that problem. We’re a small, high-impact group operating at the frontier of autonomous vehicle safety, working across probabilistic risk assessment, simulation, autonomy performance, statistical modeling, and rare-event estimation. Every teammate owns a significant piece of the overall safety story. This role is one of those pieces
• Find the edge cases before the road does. Identify failure modes and edge cases that expose weaknesses in our system before they appear in on-road environments. Write C++ to efficiently search high-dimensional failure spaces at scale
• Build probabilistic risk models. Write Python to estimate autonomy-level risk using Bayesian models inside our Probabilistic Risk Assessment framework. Your outputs will inform engineering priorities across the company
• Drive safety-informed design decisions. Provide rigorous analysis to support complex autonomy system design trade-offs that affect both safety and performance
• Run and analyze simulations. Support development of simulation scenarios, structured track testing, on-road testing, and hardware-in-the-loop test rigs and communicate risk-informed findings to engineering and leadership
• Pioneer new safety methods. When existing approaches fall short of our safety targets, you’ll develop new ones. That’s not a figure of speech, it’s a regular part of this job

Benefits

• Remote-friendly work environment
• 401k
• Generous PTO policy
• Life insurance
• Assortment of medical (PPO/HMO/HDHP), dental, vision, and FSA plans
• Family-friendly company events
• Dog-friendly office
• Employee-driven fitness classes
• Beautiful, renovated facilities in Mountain View, CA and Lancaster, TX
• Free EV charging
• Trivia and board game nights- Strong written and verbal communication — you’ll explain complex risk tradeoffs to engineers, leads, and executives, and author formal safety-case deliverables
• Production software development in Python (numpy/scipy/pandas) and modern C++ (C++17/20) — templates, numerical linear algebra, a compiled build system
• Experience producing and reviewing analysis, code, and artifacts authored by generative AI tools
• Statistical model building and validation: fitting parametric models to messy operational data, and rigorously assessing fit — with particular attention to the tails, where a good average fit can still be badly wrong
• Quantitative risk and reliability modeling of complex systems: decomposing system-level risk into estimable contributions, propagating uncertainty through that decomposition, and identifying which contributors dominate the result
• Working knowledge of tail risk metrics and how estimator uncertainty propagates into a decision threshold
• M.S. or Ph.D. in engineering, mathematics, statistics, or a related field
• Bayesian methods for sparse-evidence problems: combining prior engineering knowledge with limited field and simulation evidence, and defending a quantitative conclusion when the event of interest has never been observed
• Numerical robustness instincts: numerical stability and conditioning, transformations between sample and physical spaces, optimization, root-finding
• Reproducibility and auditability discipline — deterministic results, regression testing against known-good outputs, traceable provenance for every input assumption, and a strong preference for failing loudly over failing silently. Our outputs gate driverless deployment decisions and must regenerate to identical numbers
• Deep applied probability and statistics, including fluency with skewed and heavy-tailed distributions — not just the Gaussian/Poisson/binomial core
• Estimating the probability of events far rarer than you can directly observe or simulate — you understand why naïve sampling fails in this regime, and you’ve used advanced sampling or variance-reduction techniques to get usable estimates with quantified confidence
• Modeling time-correlated or statistically dependent stochastic processes, and reasoning about dependence structure rather than assuming independence
• Working in high-dimensional stochastic spaces, including reducing dimensionality while preserving the behavior that matters
• Systematically searching for failure conditions in cyber-physical systems — adversarial search, stress testing, or formal-methods-adjacent approaches
• Building or validating simulation environments; simulation-to-real validation
• GPU-accelerated numerical computing (CUDA and associated libraries)
• Distributed compute at scale: AWS, infrastructure-as-code (Terraform), and large-scale result storage and query (Elasticsearch or comparable)
• Hazard analysis and safety-critical standards: ISO 26262, ISO 21448 (SOTIF), UL 4600, IEC 61508, ARP4754A/ARP4761; FMEA/FMECA; formal risk-acceptance frameworks; PRA practice from nuclear, aerospace, or process industries
• Vehicle dynamics, collision kinematics, or crash severity modeling
• AV stack fluency: sensor fusion and tracking, perception error characterization, motion planning, control
• LaTeX for technical deliverables; data visualization