Not Another Humanoid: A Purpose-Built Home Robot, Trained on 500 Real Families
Sunday, a robotics startup founded by Stanford PhDs Tony Zhao (CEO) and Cheng Chi (CTO), has officially launched from stealth with Memo — a personal home robot designed not to entertain, but to perform real chores: loading dishwashers, folding laundry, tidying rooms, and brewing coffee.
Unlike industrial or humanoid robots retrofitted for the home, Memo is engineered from the ground up for domestic environments, with two foundational differentiators:
Training on 10 million real-world household task episodes, collected from over 500 actual homes
Data capture via the proprietary “Skill Capture Glove” — a wearable sensor system that records human motion, force, timing, and decision-making during routine chores
This approach bypasses the industry’s core bottleneck:
Most home robots fail not due to hardware, but because they’re trained in sterile labs — not messy, unpredictable living spaces.
Why the Data Matters: From “Demo” to “Daily Use”
Traditional home robots rely on:
Scripted demonstrations
Simulated environments
Narrow, pre-defined tasks
Memo’s training data is fundamentally different:
10 million task episodes — including multi-step sequences like “clear table → scrape plates → load dishwasher → start cycle”
High variability: different homes, lighting, clutter levels, object types
Human nuance: how people adjust grip for wet dishes, fold wrinkled shirts, or navigate around pets
This enables “long-horizon task execution” — a technical term for completing complex, multi-stage chores without human intervention.
“The problem has always been data,” said CEO Tony Zhao. “Our Skill Capture Glove changes this by collecting thousands of hours of daily routines from real families.
Design Philosophy: Safety Over Spectacle
Memo does not mimic human form. It embraces a wheeled, low-center-of-gravity design optimized for home safety:
Rolling base: Eliminates fall risk; stable even during power loss
Silicone-clad exterior: Soft, non-threatening, child- and pet-friendly
Compact footprint: Fits in standard kitchens and hallways
No exposed joints or pinch points: Meets consumer safety standards for household appliances
This is not a compromise. It is a deliberate rejection of bipedal form factors for domestic use — where reliability trumps anthropomorphism.
Commercial Path: Founder’s Beta, Not Pre-Orders
On November 19, 2025, Sunday will open applications for the Founding Family Beta:
50 households selected globally
Late 2026 delivery of individually numbered Memo units
Direct engineering access, feedback loops, and influence over feature development
This is not a crowdfunding campaign. It is a controlled, data-rich field trial — designed to refine both hardware and AI before commercial scale.
Investors see this as the right pace.
“The promise of AI robotics isn’t back-flipping demos,” said Eric Vishria, General Partner at Benchmark.
“It’s robots that work in messy, real-world situations. Tony and Cheng’s approach finally makes collecting robot-ready data at massive scale possible.”
Investment Takeaway: A New Data Moats in Consumer Robotics
Sunday’s strategy flips the traditional robotics model:
Traditional Approach
Sunday’s Approach
Build robot → collect limited data → deploy
Collect massive real-world data → build robot
Sim-to-real transfer (high failure rate)
Real-to-real learning (high reliability)
Humanoid form (high cost, low stability)
Purpose-built form (low cost, high safety)
Sell to early adopters
Co-develop with families
The result is a data moat that is hard to replicate:
Requires months of in-home data collection
Depends on proprietary capture hardware (Skill Glove)
Benefits from network effects: more homes = better generalization
While competitors chase viral demos, Sunday is building the first home robot trained on the only dataset that matters: real life.
This isn’t about when robots will enter the home. It’s about when they’ll finally work there.
Sunday’s answer: late 2026 — for 50 families. Then, the world.
