Senior AI Engineer Self-Supervised Learning

Job description

RIVR, part of Amazon is a robotics company pioneering Physical AI through real-world doorstep delivery. Founded in 2024 as an ETH Zurich spin-off, RIVR developed wheeled-legged robots designed to operate in complex, unstructured environments such as stairs, gates, doors, and uneven urban terrain. We believe that achieving general physical intelligence requires solving real customer problems in the real world, where robots can learn from rich operational data at scale. Following our acquisition by Amazon in March 2026, we are continuing this mission with greater reach and speed. By combining custom robot hardware, onboard autonomy, and cloud-based coordination, RIVR, part of Amazon is building the next generation of safe, reliable autonomous robots for last-mile delivery **Job Description** Our global fleet of autonomous robots operates in the real world, generating vast amounts of multi-modal sensor data. While our VLA team focuses on building large-scale models to consume this data, much of it remains unlabeled and unstructured. We are seeking an expert in self-supervised and representation learning to unlock the full potential of this massive data pool. In this role, you will be responsible for designing and building the core data engine that transforms raw, real-world sensor data into high-signal, structured datasets suitable for training neural networks. You will pioneer methods to automatically curate, filter, and pseudo-label this data, creating powerful representations that serve as the foundation for all downstream tasks, including navigation, imitation learning, and decision-making. You will work directly with the VLA and Reinforcement Learning teams to define data strategies and interfaces, ensuring the data you produce directly accelerates their model development. If you are passionate about solving the "data bottleneck" in robotics and want to build the systems that learn meaningful patterns from the physical world, we invite you to join us. ### What you’ll be doing - Design, build, and maintain scalable data pipelines to process, filter, and transform terabytes of raw, multi-modal sensor data (e.g., video, LiDAR, IMU, odometry) from our robotic fleet. - Develop and implement state-of-the-art self-supervised and representation learning algorithms to automatically extract features, discover patterns, and generate pseudo-labels from our unlabeled data. - Collaborate closely with the VLA Foundation Model and RL teams to define data requirements, APIs, and strategies for leveraging curated datasets and learned representations. - Architect and implement robust evaluation strategies, benchmarks, and datasets to rigorously track the performance and quality of both the data pipeline and the downstream models that consume it. - Own the data integration workflow, creating efficient data loaders and access patterns to make high-signal data readily available for model training and experimentation. - Research and prototype novel techniques in data curation, active learning, and anomaly detection to continuously improve the quality and efficiency of our data engine. ### What you must have - Master’s degree or higher in a relevant field such as Computer Science, Machine Learning, or Robotics. - A minimum of three years of industry or research experience, with PhD experience applicable. - Deep expertise in self-supervised learning (SSL) and representation learning, particularly with multi-modal sensor data (e.g., contrastive learning, masked autoencoders, world models). - Proven experience in building and managing large-scale data processing pipelines for machine learning (e.g., using Spark, Kubeflow, or similar cloud-native tools). - Strong understanding of robotic sensor data (e.g., camera, LiDAR, IMU, odometry) and their characteristics. - Strong programming skills in Python and deep experience with PyTorch, including creating custom and efficient DataLoaders. - Experience with MLOps best practices and data versioning tools (e.g., DVC, Pachyderm) ### Get some bonus points - PhD degree in Robotics, Engineering, Computer Science, Machine Learning or a similar discipline, or an equivalent amount of research experience. - Publications at top-tier ML or robotics conferences (e.g., NeurIPS, ICML, CVPR, CoRL, ICLR). - Experience with generative models (e.g., GANs, Diffusion Models) for data augmentation or simulation.