Sunday Robotics

Sunday Robotics - Memo Robot and the Zero Robot Data Approach

Sunday Robotics

Home > Companies > Sunday Robotics


Overview

Sunday Robotics is a startup co-founded by Stanford PhD graduates Tony Zhao (author of the ACT paper) and Cheng Chi (author of Diffusion Policy). They are gaining attention for their innovative approach of training without robot data, using only human motion data.

ItemDetails
HeadquartersSan Francisco, CA
Founded2024
Co-FoundersTony Zhao (CEO), Cheng Chi (CTO)
Funding$35M (Benchmark, Conviction)
RobotMemo

Founding Team

Tony Zhao (CEO)

  • Stanford PhD
  • Author of ACT (Action Chunking with Transformers) paper
  • Developer of ALOHA robot system

Cheng Chi (CTO)

  • Stanford PhD
  • Author of Diffusion Policy paper
  • Expert in robotics AI learning

Memo

Hardware Specs

ItemSpecs
TypeTorso-style, wheeled base
Height1.7m
Weight77kg
Arms2× multi-DoF
GripperDual
Height AdjustmentFloor to 2.1m

Capable Tasks

  • Dish organization and dishwasher loading
  • Handling wine glasses (delicate objects)
  • Espresso brewing
  • Laundry folding

Zero Robot Data Approach

Skill Capture Gloves

Using $200 gloves instead of robot teleoperation:

ItemTeleop RigSunday Gloves
Cost~$20,000~$200
LocationLabReal homes
CollectorsExperts”Memory Developers”
DataRobot motionsHuman motions

Data Pipeline

[Human wearing gloves] → [Performing chores at home] → [Motion data collection]

                                                  ┌───────────▼───────────┐
                                                  │   Skill Transform     │
                                                  │ (Human→Robot motion)  │
                                                  └───────────┬───────────┘

                                                  ┌───────────▼───────────┐
                                                  │       ACT-1 Model     │
                                                  └───────────────────────┘

Data Scale

ItemValue
Gloves Distributed2,000+
Data Collection Homes500+
Household Task Episodes10M+

ACT-1 Model

Features

ItemDetails
Core InnovationZero Robot Data
Training DataHuman motions only
Combined FeaturesManipulation + Navigation
GeneralizationZero-shot to new environments

Skill Transform

Software for converting human motion → robot motion:

  • Claims 90% success rate
  • Human dexterity → Robot gripper mapping
  • Includes force/contact data

Deployment Plans

TimelinePlan
2025.11Memo launch, beta applications begin
Late 2026”Founding Family” beta (~50 households)
Current Production Cost~$20,000 (handmade)
Target Retail Price<$10,000 (at scale)

Significance

If Sunday’s approach succeeds:

ProblemSolution
Data BottleneckLarge-scale collection via affordable gloves
Environment DiversityCollection in real homes
CostHome robots under $10,000
GeneralizationZero-shot adaptation to new environments

References


See Also