Sunday's 10 Million Trajectories Lead

The key unlock is that Sunday is treating home robotics like a data collection problem before it treats it like a robot deployment problem. Ten million trajectories means it can train ACT-1 on thousands of examples of opening dishwashers, placing glasses, and folding laundry without buying, maintaining, or teleoperating a fleet in real homes. That sharply lowers the cost of learning and gives Sunday a wider range of household edge cases earlier.

• Most robot companies gather training data by running actual machines in factories or homes, often with teleoperation. Sunday instead uses gloves and a head camera in 500 plus households, which is closer to how autonomous driving first scaled with huge sensor datasets before full autonomy worked reliably.
  1 sacra 2 sacra
• The comparison set shows why this matters. 1X and Figure are building humanoids that can in principle do more tasks, but they also need expensive hardware in the loop to collect real world interaction data. Sunday can generate manipulation data before broad robot rollout, which changes burn and speed.
  1 sacra 3 sacra 4 sacra
• A trajectory is one full recorded action path, like reaching for a plate, gripping it, moving to the dishwasher, and placing it correctly. At 10 million examples, the model is not learning a single perfect motion, it is learning the many slightly different ways the same chore gets done in messy homes.
  1 sacra 2 sacra

The next step is turning this data advantage into a product advantage. If Sunday keeps converting glove demonstrations into robot actions with high fidelity, it can reach homes with a cheaper, narrower robot that works well on common chores, then use live deployments to compound an already large training lead.