Senior Machine Learning Engineer, Weather & Degraded Road Surfaces

Waymo•San Francisco, CA

1d•Hybrid

About The Position

Waymo is an autonomous driving technology company focused on building the Waymo Driver to improve mobility and safety. The Perception team develops the system that interprets the surrounding environment of the autonomous driving vehicle (ADV). This involves research, development, and collaboration with other teams within Waymo and Alphabet. The team has access to extensive driving data to develop methods for learning from large-scale data, building and training models, analyzing real-world behavior, and optimizing models for hardware. The DRAW (Degraded Road Surfaces and Weather) team specifically addresses challenges related to adverse weather and road conditions. Current work includes large supervised multi-model models and few-shot detection using Vision Language Models (VLMs). Past projects include developing ML weather estimators, sensor cleaning signals, and models to avoid floods, puddles, potholes, sand, and debris. The new focus is on snow, including understanding road friction, snow accumulation, and tire tracks. In this hybrid role, you will report to a Technical Lead Manager and apply machine learning techniques to develop multi-modal sensor fusion architectures and spatial-temporal representation learners for various perception tasks. You will also develop scalable training methods for large datasets and models on Alphabet's compute infrastructure, including pre-training, post-training, distributed fine-tuning, and regression avoidance. Additionally, you will develop and maintain model evaluation recipes and metrics.

Requirements

Bachelors in Computer Science or a similar discipline, or an equivalent amount of deep learning experience
5+ years experience in Machine Learning and/or Computer Vision
Experience with Python
Experience with ML frameworks like PyTorch, JAX, or Tensorflow.

Nice To Haves

MS or PhD Degree in Machine Learning, Robotics, Computer Science or a similar discipline
Publications at top-tier conferences like CVPR, ICCV, ECCV, ICLR, ICML, ICRA, IROS, RSS, NeurIPS, AAAI, IJCV, PAMI
Experience with C++

Responsibilities

Apply machine learning techniques to build multi-modal sensor fusion architectures and spatial-temporal representation learners for object detection and tracking, occupancy and semantic segmentation, road understanding, etc.
Develop scalable recipes for large data, large model training running on Alphabet’s compute infrastructure, create methods and recipes for pre-training and post-training.
Develop methods and recipes for distributed fine-tuning enabling multiple developers to simultaneously improve the model, develop methods and recipes to avoid regression against a production system.
Develop and maintain model evaluation recipes and metrics for measuring and improving performance of pre-trained and fine-tuned models