Memo: Home Robot Trained on 10M Episodes

24 Nov 2025

Sunday Robotics — freshly out of stealth with $35M from Benchmark and Conviction — unveiled Memo, a home robot trained on 10 million episodes of real family routines. The team captured everyday chores (clearing tables, loading dishwashers, folding laundry, pulling espresso shots) using $200 “Skill Capture” gloves that mirror Memo’s hand geometry and sensors — a far cheaper alternative to traditional teleoperation setups (which the piece cites at ~$20K each). If a human can perform a task wearing the glove, Memo can learn it.




Memo’s demo performance is striking: the table-to-dishwasher run involves 68 dexterous interactions with 21 objects over 130+ feet of navigation, and the team reports zero wine-glass breakages across 20+ live demos. The data set even included real-world chaos — cats in dishwashers, plums on tables — which the company argues helps robustness. The founding team includes Stanford robotics PhDs Tony Zhao and Cheng Chi, several ex‑Tesla FSD engineers, and an undergrad (Alper Canberk) training models. Beta applications are open for 50 founding families; shipments are planned for late 2026.