Sunday ACT-1
Sunday Robotics' Zero Robot Data Based Foundation Model
Sunday ACT-1
Home > Models > Sunday ACT-1
Key Significance
- Zero Robot Data: Trained purely on human movements (glove data) without any robot data - paradigm shift in data collection
- $200 Gloves vs $20K Teleop: 100x cheaper data collection cost enabling scalability
- Large-Scale Real Home Data: “Memory Developers” collected 10M+ episodes from 500+ homes - real environments, not labs
- Zero-shot Generalization: Generalization capability to work in previously unseen homes
- Consumer Price Target: Target <$10,000 with mass production - making home robots realistic
- Long-Horizon Manipulation + Navigation Integration: First model handling both manipulation and map-based navigation in single end-to-end model
- Complex Real Tasks: In Table-to-Dishwasher, traveled 130+ feet, 33 interactions, handling 21 objects
Overview
ACT-1 is a foundation model developed by Sunday Robotics, trained without any robot data using only human movements. It is the first model combining long-horizon manipulation and map-based navigation in a single end-to-end model.
Key Innovation: Zero Robot Data
Data Collection Method
Using $200 gloves instead of traditional teleoperation:
| Item | Teleop Rig | Sunday Gloves |
|---|
| Cost | ~$20,000 | ~$200 |
| Location | Labs | Real homes |
| Collectors | Experts | ”Memory Developers” |
Data Scale
| Item | Value |
|---|
| Gloves Deployed | 2,000+ |
| Data Collection Homes | 500+ |
| Chore Episodes | 10M+ |
Table-to-Dishwasher Task
“The most complex task autonomously performed by a robot”:
| Item | Value |
|---|
| Autonomous Navigation | 130+ feet |
| Unique Interactions | 33 |
| Number of Objects | 21 |
| Included Objects | Delicate dishes |
Dexterous Manipulation
- Two wine glasses in one hand (without breaking)
- Wine glass insertion into dishwasher
- Picking up 2 utensils with one hand
Zero-shot Generalization
Emphasizes generalization capability to work in previously unseen homes.
Hardware: Memo
| Item | Spec |
|---|
| Height | 1.7m |
| Weight | 77kg |
| Form | Torso style, wheel base |
| Arms | 2x multi-DoF |
| Grippers | Dual |
| Height Adjustment | Floor ~ 2.1m |
Availability
| Item | Details |
|---|
| Current | Pre-commercial |
| 2026 | ”Founding Family” Beta (~50 households) |
| Current Build Cost | ~$20,000 (handmade) |
| Target Retail Price | <$10,000 (mass production) |
Comparison with Other Approaches
| Approach | Advantages | Disadvantages |
|---|
| Teleoperation | Accurate robot data | Expensive, slow |
| Simulation | Large-scale generation | Sim-to-Real Gap |
| Human Video | Cheap | Action extraction difficult |
| Sunday (Gloves) | Cheap, real environment, high fidelity | Scale validation needed |
Significance
If Sunday’s approach succeeds:
- Solves Data Bottleneck: Cheap in-home data collection
- Zero-shot Generalization: Adapts to new environments
- Consumer Price Point: Home robots under $10,000
References
See Also