MLOps Engineer - Humanoid Robot Data Project

About the role

The AI Robot Association (AIRoA) is running a national project to collect one million hours of humanoid robot operation data and use it to train large Vision-Language-Action (VLA) models. The project is supported by a substantial investment of 20.5 billion yen from NEDO. The MLOps Engineer works in the Development Division, designing and maintaining the ML infrastructure and pipelines used by VLA researchers.

Key responsibilities

• Design, implement and maintain large-scale ML pipelines and optimise model performance for training on massive robot datasets.
• Design, deploy and maintain distributed training clusters to shorten model development cycles.
• Work closely with VLA researchers to capture requirements for data preprocessing, training, monitoring, evaluation and deployment, and continuously improve the ML pipeline.
• Optimise ML infrastructure and pipelines for cost, performance and reliability.
• Build and maintain MLOps tools so researchers can visualise and analyse model performance.

Required profile

• Master's degree in Computer Science, Engineering or a related field (or equivalent practical experience).
• 5+ years of professional experience as a software engineer in MLOps, machine learning or robotics.
• Experience deploying distributed systems on cloud services such as AWS, GCP or Azure.
• Experience with orchestration tools such as Airflow, Dagster or Kedro.
• High proficiency in PyTorch or JAX.

Preferred qualifications

• Training and fine-tuning of RL, VLM and VLA models, including distillation, supervised fine-tuning and policy optimization.
• ETL/ELT pipelines, hyperparameter optimization and distributed training frameworks.
• GPU memory management, data quality monitoring, and robotics sensor data processing (RGB/depth images, point clouds).
• ROS/ROS2, SQL, performance analysis tools, and business-level English.

The work location is the Tokyo Ryutsu Center in Ota-ku, Tokyo.