Macrodata

Macrodata Labs is a robotics-data infrastructure startup that helps robotics teams transform raw robot and egocentric video recordings into training-ready datasets. Its pipelines segment long recordings into timestamped subtasks, recover metric 3D hand motion from moving cameras, assess whether footage contains the signal a team needs, and estimate task progress and reward signals for reinforcement learning. The founding team previously worked on widely used open LLM pre-training datasets before turning to the emerging bottleneck in physical-world AI: robot fleets and wearable cameras now produce more footage than teams can feasibly review by hand. Macrodata also maintains Refiner, an open-source framework for reading, transforming, and writing robotics datasets.