Driver Refinement and Validation’s (DRV) mission is to build upon the high-quality ML Driver developed by Driver Intelligence, refining its performance to improve autonomous driving in specific, complex real-world contexts and problem areas (e.g., Ride-Hailing contexts). It aims to ensure that the Waymo Driver always computes safe, compliant, comfortable, and useful plans. As a Machine Learning Engineer in DRV, you will bridge the gap between cutting-edge machine learning and real-world vehicle performance, engineering the validation and onboard systems that ensure the autonomous vehicle reliably executes intelligent and predictable maneuvers.

You will:

• Own ML features for Scene Understanding (traffic controls, signs, special regions such as RRX and flood zones, lane states) end to end: data → model → eval → onboard launch
• Mine and curate large-scale driving data (real and simulated) to find long-tail failures, build training and eval sets, and design labeling and supervision
• Design, train, and iterate on model tasks and heads in large multi-task models. Improve calibration, robustness, and generalization to new geographies
• Build rigorous evaluation: PRT metrics, slicing taxonomies, side-by-sides, and staging launches, and turn the results into launch decisions
• Integrate models into the onboard stack (C++), combining learned outputs with safe heuristic fallbacks, Mendel-guarded rollouts, and map-contradiction handling
• Triage real driving events, find the root cause anywhere in the stack, and propose surgical or systemic fixes. Work closely with Perception, Reasoning, Mapping, and ML Infrastructure teams

You have:

• BS/MS/PhD in CS, EE, Robotics, ML, or a related field, or equivalent practical experience.
• Hands-on experience shipping ML models to production, including training, offline evaluation, and deployment (L4: 2+ yrs; L5: 4+ yrs, or a PhD plus 2+ yrs)
• Strong data mining / data engineering: building large-scale batch pipelines (Beam/Flume/Spark or similar) and analytical SQL to find, slice, and curate data
• Proficiency in Python with a modern DL framework (JAX, TensorFlow, or PyTorch) and production C++
• Demonstrated cross-stack generalist work: has owned problems that crossed data, model, and systems/serving boundaries
• Rigorous evaluation instincts: metric design, error analysis and slicing, and A/B or staged rollout

We prefer:

• AV/Robotics experience, especially perception (detection, classification, tracking)
• Experience with transformer or temporal models (GRU/RNN), multi-task learning, and model calibration/uncertainty
• Labeling-pipeline and supervision design; autolabeling or synthetic data
• TPU/accelerator performance work; exporting models for onboard/edge inference
• Builds internal tools (visualization, triage, experiment tracking) that accelerate the team