AI Engineer

The AI Robot Association (AIRoA) is collecting one million hours of humanoid robot operation data to train Vision-Language-Action (VLA) models in an open robot data ecosystem. This AI engineer develops the multimodal models and the training systems around them.

The engineer builds Vision-Language or comparable multimodal models with robotics applications in mind, fine-tunes and evaluates existing models and analyzes errors. The work includes training pipelines on real-world data, evaluation metrics and iterative improvement cycles, preparing and preprocessing image, video, language and action data, building training environments, and following new research to bring it into model improvements. The role also sets up training and inference infrastructure.

Applicants need experience taking machine learning models from deployment through improvement and operation in a production service, implementing, training and evaluating models in Python and PyTorch, and hands-on fine-tuning of Vision-Language or similar multimodal models. Experience with pipelines, evaluation design and error analysis on real data is also required. VLA or robotics experience, ROS, Sim2Real, cloud infrastructure and MLOps are preferred.