Imagine building a dataset for robots that need to operate in real homes. You would need more than clean images of furniture. You would need different room layouts, cluttered surfaces, changing light, objects partially hidden from view, and the many small variations people create every day. That is what makes real-world data collection challenging. Useful datasets must represent the environments where AI systems will actually be used. I'm interested in how @vangrid_io fits into the broader effort to connect real-world capture with the needs of Physical AI development. If you were designing a household robotics dataset, which difficult scenario would you make sure to include first?