Sunday Robotics debuts Memo, a wheeled home robot trained via data-capture glove from 500+ homes
Nov 20, 2025 with Tony Zhao
Key Points
- Sunday Robotics raises $29.5 million Series A led by Andreessen Horowitz to scale Memo, a wheeled home robot trained exclusively on human motion data captured through a proprietary glove rather than teleoperation or simulation.
- The startup plans to launch a beta program in late 2026 and general availability in 2027 or 2028, targeting home tidying tasks like picking up objects and sorting items as its initial use case.
- Co-founder Tony Zhao, who left DeepMind, Tesla, and Google in early 2024, argues that decoupling data collection from robot deployment via the glove lets Sunday scale training faster than competitors relying on deployed robots in the field.
Summary
Tony, co-founder and CEO of Sunday Robotics, is building Memo, a wheeled home robot trained entirely on human motion data captured through a proprietary skill-transfer glove. The company is a year and a half old, founded by two former Stanford robotics PhD students — Tony previously worked at DeepMind, Tesla, and Google before leaving academic research in early 2024 to build a product rather than publish papers.
The data strategy
Memo's training relies exclusively on glove-captured human demonstrations. Sunday does not use teleoperation, simulation, or world models for specific behaviors — only internet-scale pretraining for general knowledge, with all task-specific skills learned from glove data. Tony argues teleoperation alone can't scale fast enough: autonomous driving took Tesla nearly a decade with millions of cars collecting data daily, and robotics is a harder problem. The glove lets Sunday decouple data collection from robot deployment entirely — humans wear the glove, capture the behavior, and Memo learns it without a robot ever needing to be in the field first.
The scale question is open. The path to a large training corpus isn't 80 million gloves — Tony frames it more like the LLM compute flywheel: the product generates revenue, revenue funds more data collection, and the model improves. Sunday doesn't need to solve general robotics to ship; home tasks are low-stakes and relatively simple, but carry real emotional and functional value for users.
Form factor and design
Memo is wheeled, not legged. The design decision centers on passive safety: if the robot is fully extended and power is cut, it stays stable rather than toppling. A legged robot can't offer that guarantee. Wheels also simplify battery management and docking. The camera sits beneath the head rather than in the eyes — a deliberate choice to avoid the unsettling effect of making eye contact with a lens.
The near-term use case Tony points to is tidying: picking up objects off the floor, sorting items, handling wine glasses. Not full dishwasher loading, but enough to have clear willingness to pay if the price is right.
Timeline and funding
Sunday's beta program is planned for late 2026, placing Memo in real homes to understand user behavior. General availability is targeted for 2027 or 2028 depending on beta results. Tony announces a Series A of $29.5 million led by Andreessen Horowitz, with Accel and Coastal Ventures also participating.
The segment also briefly covers Flexion, a Zurich-based robotics intelligence company founded by Nikita, which takes a contrasting approach: simulation-first training using reinforcement learning, layered with LLMs for common-sense reasoning and semantic segmentation models to reduce the visual complexity that makes sim-to-real transfer hard. Flexion announces a $50 million raise with participation from DST Global, Nvidia Ventures, Promus Ventures, and Moonfire. Nikita is in San Francisco scouting for a second office while the core team remains in Switzerland.