Sunday Trains Memo Without Teleoperation

Sunday is betting that the scarce resource in home robotics is not robots, it is diverse household behavior data. By putting a $400 glove and camera kit in many homes instead of putting expensive teleoperated robots in a few homes, Sunday can collect chore demonstrations across more layouts, objects, and edge cases at far lower cost. That makes Memo look less like a one off hardware demo and more like a model trained on the messiness of everyday homes.

• This changes the economics of data collection. Sunday says the glove setup mirrors Memo’s 3 finger gripper and converts human motions into robot trajectories at about 90% fidelity, which lets it build training data without shipping, maintaining, and remotely operating full robots in customer homes.
  1 sacra
• The main alternative is teleoperation. 1X trains Redwood on teleoperated and autonomous robot episodes and is hiring operators for robot sessions and eventual customer homes. Figure is gathering human video across Brookfield’s 100,000 residential units. Sunday is choosing the cheapest path to volume, even if translation from human hands to robot actions is imperfect.
  2 sacra 4 1x 5 1x 6 figure
• That approach fits Memo’s form factor. A wheeled robot that does table clearing, dishwasher loading, and laundry does not need to copy the full human body. It needs enough hand and camera data to learn where objects are, how fragile they are, and what multi step household tasks look like across unfamiliar rooms.
  1 sacra 3 sacra

The next phase is turning data advantage into deployed advantage. If Sunday can keep improving Skill Transform and funnel glove data into reliable zero shot performance, it can reach homes and managed properties with a lower cost robot and a larger pretraining set than humanoid rivals that depend on slower, more expensive teleoperated collection.