Robotics Software Engineer, Mission Evaluation & Fine-Tuning

Job description

[FieldAI](http://www.fieldai.com)’s Irvine team is where embodied AI meets real robots, real sensors, and real field deployments. Based in the heart of Southern California’s robotics ecosystem, we build risk-aware, reliable, field-ready AI systems that solve the hardest problems in robotics and unlock the full potential of embodied intelligence. If you want your work to ship, get tested on hardware, and improve through real deployments, Irvine is the place. We go beyond typical data-driven approaches or pure transformer-only architectures, combining rigorous engineering with learning systems proven in globally deployed solutions that deliver results today and get better every time our robots run in the field. ### What You Will Get To Do - Own the end-to-end inspection behavior success metric. - Build the evaluation loop: golden-labeled datasets, seeded-fault runs, and regression suites at feature, system, and release level that gate every release candidate. - Turn field evidence into training and evaluation sets. - Map technical performance to the metrics the customer measures. - Make failures legible: triage mission logs, attribute each miss to navigation, localization, perception, or behavior, and route it to the owning team with evidence. ### What You Have - MS in computer science, robotics, aerospace engineering, or a related field, with 3+ years in robotics or ML engineering. - Strong Python, working C++, and comfort with ROS 2 logs, bags, and replay tooling. - Experience evaluating and fine-tuning vision or vision-language models (PyTorch, LoRA or similar), including building the datasets and metrics behind them. - Statistical rigor: knowing the difference between a benchmark number and a field result, and how to design an experiment that tells them apart. - Data pipeline skills to move evidence from robot to training set to dashboard. ### What Will Set You Apart - Authored or maintained a robotics benchmark, evaluation suite, or challenge. - Simulation-based evaluation and sim-to-real validation. - MLOps for on-edge models: quantization, versioning, and rollout to a fleet. - Industrial inspection, gauge reading, or anomaly detection experience.