Sascha Kaczmarek

Topic: "Real-World Work Data: The Missing Data Layer for Physical AI?"

January 20, 2027 · Time TBA · Main Stage

For Physical AI, it is not enough to simulate the physical world or train robots exclusively on synthetic data and teleoperation. AI systems must understand how humans act in real environments, make decisions, and execute complex work processes.

Especially in production and logistics, a previously overlooked data source emerges: Real-World Work Data. Movements, spatial context, process steps, interactions with tools and machines, as well as human decisions together form a multimodal representation of real work. This data can form a critical bridge between human expertise, digital simulation, and future robotic systems.

The talk shows how such data can be captured, structured, and made usable for training Physical AI in real production and logistics environments. Topics include the challenges of scaling, annotation, data quality, generalizability, and transferring human movement and action patterns to different robotic systems.

Based on initial pilot projects, the talk also discusses what insights can already be derived from combining real movement data, spatial information, digital twins, simulation, and robotic systems. What data does Physical AI actually need in order to not only understand the world, but to act within it?

The talk thus raises the central question of whether real human work represents an independent and previously underestimated data layer for the development of the next generation of Physical AI and humanoid robots.