# Sunday Robotics

> Source: https://aiwiki.ai/wiki/sunday_robotics
> Updated: 2026-06-24
> Categories: Humanoid Robots, Robotics Companies
> From AI Wiki (https://aiwiki.ai), a free encyclopedia of artificial intelligence. Quote with attribution.

**Sunday** (often referred to as **Sunday Robotics**, stylized as **sunday.ai**) is an American robotics company in [Mountain View, California](/wiki/mountain_view_california) that is building [humanoid robot](/wiki/humanoid_robot)s for the home, beginning with a wheeled domestic assistant named [Memo](/wiki/memo_robot) that learns chores from human demonstrations rather than factory teleoperation. Co-founded in 2024 by Stanford-trained roboticists Tony Zhao (CEO) and Cheng Chi (CTO), the researchers behind [ALOHA](/wiki/aloha) and [Diffusion Policy](/wiki/diffusion_policy), Sunday emerged from stealth on November 19, 2025 and reached a $1.15 billion valuation in March 2026 on a $165 million Series B led by [Coatue](/wiki/coatue_management). Its central bet is that the bottleneck for home robots is not algorithms or hardware but data, and that a roughly $200 wearable glove can gather real household [imitation learning](/wiki/imitation_learning) data far more cheaply than any teleoperation rig.[1][2][3]

Sunday's founders' academic work on ALOHA, [Action Chunking with Transformers](/wiki/action_chunking_with_transformers) (ACT), and Diffusion Policy underpins much of the modern wave of imitation-learning robotics. Memo, a wheeled robot designed to fold laundry, clear tables, load dishwashers, and pull espresso shots, was revealed alongside Sunday's ACT-1 foundation model and a "Founding Family Beta" program. The Series B, raised roughly two years after the founders started prototyping in a garage, made the company a unicorn and shifted its focus from demos to real-world home deployment in late 2026.[1][2][3]

## History

### Who founded Sunday Robotics?

Sunday Robotics was founded in 2024 by Tony Zhao and Cheng Chi, who had been pursuing parallel research threads at [Stanford University](/wiki/stanford_university). The two first connected on X (formerly Twitter) after coming across each other's papers, which had been published roughly a month apart. Both were already known inside the robot-learning community: Zhao for [ALOHA](/wiki/aloha) and Mobile ALOHA at Stanford, and Chi for [Diffusion Policy](/wiki/diffusion_policy) (at [Columbia University](/wiki/columbia_university) and Stanford) and the [Universal Manipulation Interface](/wiki/universal_manipulation_interface) (UMI).[2][4][5]

The two left their PhD programs to start the company. Sarah Guo and Pranav Reddy of the venture firm [Conviction](/wiki/conviction) provided what Sunday describes as its first term sheet around the spring of 2024. The company then spent roughly 18 months in stealth out of a garage, running 3D printers around the clock to iterate on Memo's hardware while building its data-collection and training stack.[2][6]

### What did Sunday build during its stealth period (2024 to 2025)?

During stealth, Sunday assembled a team drawn from the most active groups in robot learning and autonomy. Public reporting documented hires from [Tesla](/wiki/tesla_inc)'s Autopilot, Optimus, and AI infrastructure teams, including Nishant Desai (ML for FSD and Autopilot), Nadeesha Amarasinghe (former engineering lead for AI infrastructure), and Perry Jia (Optimus and Autopilot data engine, now head of data operations at Sunday). Other team members came from [DeepMind](/wiki/google_deepmind), [Waymo](/wiki/waymo), [Meta](/wiki/meta_platforms), [OpenAI](/wiki/openai), [Apple](/wiki/apple_inc), and [Neuralink](/wiki/neuralink). At launch in November 2025 the company had roughly 25 employees; by the Series B in March 2026 that number had grown past 70.[1][2][7]

### When did Sunday launch publicly?

On November 19, 2025, Sunday emerged from stealth with three announcements: the unveiling of the Memo robot, the publication of a technical journal post describing the ACT-1 foundation model, and the opening of applications for a "Founding Family Beta" program. Coverage in [Yahoo Finance](/wiki/yahoo_finance), [SiliconANGLE](/wiki/siliconangle), Interesting Engineering, eWeek, and Humanoids Daily framed Memo as a deliberately different bet from bipedal humanoids built by competitors such as [Figure](/wiki/figure_ai), [1X](/wiki/1x_technologies), and [Tesla](/wiki/tesla_optimus). Rather than pursuing legs and a humanlike form factor, Sunday picked a wheeled, vertically telescoping torso designed for indoor home environments.[1][8][9]

The company also disclosed roughly $35 million in early funding from [Benchmark](/wiki/benchmark_venture_capital_firm) and [Conviction](/wiki/conviction), reported by most outlets as combined seed and Series A.[1][7]

### How did Sunday reach unicorn status (March 2026)?

On March 12, 2026, Sunday announced a $165 million Series B at a $1.15 billion post-money valuation, led by [Coatue](/wiki/coatue_management) (with Coatue's Thomas Laffont joining the board) and participation from [Bain Capital Ventures](/wiki/bain_capital_ventures), [Fidelity Management & Research Company](/wiki/fidelity_investments), [Tiger Global](/wiki/tiger_global_management), Benchmark, Conviction, and Xtal Ventures. The new capital is earmarked for scaling Memo production, expanding the Founding Family Beta, and growing the team (Sunday reported tripling engineering, quadrupling research, and 5x-ing data operations headcount). By the time of the round, the company said its beta waitlist had grown to several thousand households, and it framed the raise as a shift "from demos" to bringing its first autonomous home robot into households in late 2026.[3][6][10][13]

## Who are Sunday's founders?

| Founder | Role | Background |
|---------|------|------------|
| [Tony Zhao](/wiki/tony_zhao) | Co-founder, CEO | BS, EECS, [UC Berkeley](/wiki/uc_berkeley) (2021); PhD candidate, Computer Science, [Stanford University](/wiki/stanford_university) (advisor: [Chelsea Finn](/wiki/chelsea_finn)); creator of [ALOHA](/wiki/aloha), Mobile ALOHA, and the [Action Chunking with Transformers](/wiki/action_chunking_with_transformers) (ACT) algorithm; prior internships at [Tesla](/wiki/tesla_inc) Autopilot, [Google DeepMind](/wiki/google_deepmind), and Google X / Intrinsic. |
| [Cheng Chi](/wiki/cheng_chi) | Co-founder, CTO | PhD in Computer Science, [Columbia University](/wiki/columbia_university) (advisor: [Shuran Song](/wiki/shuran_song)); visiting researcher at [Stanford University](/wiki/stanford_university); lead author on [Diffusion Policy](/wiki/diffusion_policy) (RSS 2023) and the [Universal Manipulation Interface](/wiki/universal_manipulation_interface) (RSS 2024, Best Systems Paper Finalist); prior work on Iterative Residual Policy and GarmentNets. |

Zhao's ALOHA introduced a low-cost teleoperation rig together with the ACT algorithm, which predicts short sequences ("chunks") of actions instead of single steps. Chi's Diffusion Policy reframed policy learning as a denoising diffusion process and reported an average 46.9% improvement over prior baselines across 12 tasks drawn from four robot-manipulation benchmarks. Both lines of work are now widely cited references for the imitation-learning approach Sunday is scaling.[4][5]

## Products

### What is Memo?

[Memo](/wiki/memo_robot) is Sunday's first and currently only product. It is a wheeled, two-armed household robot with a telescoping central column and a stylized cartoon face. Several reviewers, including SiliconANGLE and Interesting Engineering, compared its aesthetic to Baymax from Big Hero 6.[1][8]

| Specification | Details |
|--------------|--------|
| Height (standing) | 1.7 m (5 ft 7 in) |
| Weight (with battery) | 77 kg (170 lb) |
| Horizontal reach | 0.8 m (2 ft 7 in) |
| Vertical reach | Up to 2.1 m (6 ft 11 in) via telescoping "Z-axis" spine |
| Mobility | Wheeled mobile base |
| End-effectors | Dual grippers with 4 degrees of freedom per hand |
| Battery runtime | Approximately 4 hours per charge |
| Recharge time | About 1 hour to 80% |
| Ingress rating | IP67 (hand), IP66 (lower arm) |
| Cladding | Rigid and elastic polymers, soft silicone outer shell |
| Foundation model | ACT-1 (proprietary) |
| Build cost (hand-built) | ~$20,000 per unit (2025) |
| Target retail price | Under $10,000 at scale |

### Why did Sunday choose wheels over legs?

The wheeled base was a deliberate choice. Zhao has framed legs as the wrong constraint for an indoor product, arguing that wheeled mobility removes a class of balance failures that bipedal robots still struggle with, and that the compute spent on locomotion is better spent on dexterous manipulation. Memo also stays stable if power is lost, which avoids one of the more dangerous failure modes of bipedal humanoids in a home with children or pets.[1][8][9]

#### What can Memo do?

In launch demos and follow-up videos between November 2025 and early 2026, Memo has been shown:

- Folding and stacking socks into tucked bundles
- Loading and unloading dishwashers
- Clearing tables of plates, cups, and glassware
- Sorting and folding laundry
- Pulling espresso shots from a home coffee machine
- Handling fragile wine glasses
- Sorting pairs of shoes by the door
- Navigating unfamiliar homes (including [Airbnb](/wiki/airbnb) units used as test sites)

A reference long-horizon task in Sunday's technical materials is a single autonomous "Table-to-Dishwasher" sequence with 33 unique dexterous interactions (68 in total) across 21 objects and more than 130 feet of room-scale mobile manipulation.[11]

## Technology

### What is the ACT-1 foundation model?

Memo is controlled by **ACT-1**, an in-house robot foundation model. In a journal post titled "ACT-1: A Robot Foundation Model Trained on Zero Robot Data," published November 19, 2025, Sunday described the model as combining dexterous hand control, room-scale mobile manipulation, and map-conditioned navigation in a single end-to-end network. The name is a nod to Zhao's earlier ACT algorithm, but ACT-1 is a much larger and broader system that the company says was trained without traditional robot teleoperation data.[11]

The motivation is the cost of conventional data collection. Sunday's ACT-1 post argues, "Unless we invent a method orders of magnitude more efficient than teleoperation, gathering the data for a general-purpose robot will take decades."[11]

The core idea is what Sunday calls **Skill Transform**: a software bridge that maps the trajectories of human hands wearing the Skill Capture Glove onto Memo's matched gripper kinematics. The company's slogan is "If a human can do it in the glove, the robot can also do it." Sunday reports a roughly 90% success rate at converting glove-collected demonstrations into executable robot actions, which is why the company has been able to build a large dataset without renting expensive teleoperation rigs by the hour.[8][11]

### How does the Skill Capture Glove work?

The **Skill Capture Glove** is a wearable that mirrors Memo's hand kinematics. It costs roughly $200 per unit to build, compared with about $20,000 for a traditional teleoperation rig of comparable fidelity. Sunday has shipped over 2,000 gloves to a network of more than 500 paid contributors across the United States, internally called "Memory Developers." Each contributor wears the glove while doing chores in their own home, generating paired visual and motion data that is uploaded for training.[6][8]

### How does Sunday collect its training data?

By the November 2025 launch, Sunday had collected approximately **10 million chore episodes** from more than 500 homes, which the company described as the largest in-the-wild household dataset assembled by any robotics startup at that point. The data spans kitchens, living rooms, laundry rooms, and entryways, with deliberate emphasis on the messy parts of real homes (sock piles, cluttered counters, mismatched dishware) rather than laboratory settings.[1][8][9]

The training pipeline relies on imitation learning rather than reinforcement learning. The intuition, repeated by Zhao in interviews, is that the bottleneck for home robots has not been algorithms or hardware in isolation; it is the lack of diverse, real-world demonstration data. Sunday's bet is that solving that problem at the source, with cheap gloves and a paid contributor network, beats teleoperating robots in factories or simulating homes from scratch.[1][8]

### How does Sunday differ from Figure, 1X, and Tesla Optimus?

Sunday's approach contrasts with other foundation-model robotics companies on both form factor and data source:

| Company | Form factor | Primary training data | Notable model(s) |
|---------|-------------|----------------------|------------------|
| Sunday | Wheeled, dual-arm | Human glove demonstrations from 500+ homes | ACT-1 |
| [Physical Intelligence](/wiki/physical_intelligence_company) | Multiple embodiments | Cross-embodiment robot data | pi-0, pi-0.5 |
| [Figure](/wiki/figure_ai) | Bipedal humanoid | Mostly first-party teleoperation | Helix |
| [1X](/wiki/1x_technologies) | Bipedal humanoid | Teleoperation and in-home data | World model and Redwood |
| [Tesla Optimus](/wiki/tesla_optimus) | Bipedal humanoid | VR teleoperation and synthetic data | In-house |

Sunday sits in the same broad wave of vision-language-action robotics as Physical Intelligence, Skild AI, Figure, and 1X, and is widely seen as Conviction's flagship household-robotics bet.[2][12]

## Business model

### What is the Founding Family Beta?

Sunday's go-to-market starts with the **Founding Family Beta**. Applications opened on November 19, 2025; the company said it would pick 50 households as early adopters, each receiving a hand-numbered Memo unit and direct support from Sunday's engineering and operations team, with the program launching in late 2026. Selected beta participants are not charged, and Sunday treats their feedback and on-robot logs as a first-party data flywheel that feeds back into ACT-1. The waitlist reportedly grew into the thousands by March 2026.[1][3][6][9]

### How much will Memo cost?

A single hand-built Memo costs Sunday roughly $20,000 to produce as of 2025. The stated target retail price, once Memo is manufactured at scale, is under $10,000. The figure positions the robot as a household capital purchase rather than an enterprise device, comparable to a domestic car or high-end appliance set. Hitting that price is widely seen as the hardest engineering and supply-chain problem the company faces, and the one most directly tied to the survival of the business.[10]

## How much has Sunday Robotics raised?

| Round | Amount | Lead investor(s) | Other participants | Valuation | Date |
|-------|--------|------------------|--------------------|-----------|------|
| Seed and Series A (combined) | ~$35 million | [Benchmark](/wiki/benchmark_venture_capital_firm), [Conviction](/wiki/conviction) | (private at stealth) | Not disclosed | 2024 to November 2025 |
| Series B | $165 million | [Coatue](/wiki/coatue_management) | [Bain Capital Ventures](/wiki/bain_capital_ventures), [Fidelity Management & Research Company](/wiki/fidelity_investments), [Tiger Global](/wiki/tiger_global_management), Benchmark, Conviction, Xtal Ventures | $1.15 billion | March 12, 2026 |

The Series B is notable not just for size but for its investor mix. Coatue, Tiger Global, and Fidelity are growth-stage investors that had pulled back from hardware bets after the 2022 to 2023 correction; their participation in a household-robot company at unicorn pricing was read in the press as a signal that institutional capital is again willing to underwrite hardware-plus-foundation-model companies, at least ones with working demos.[3][6][10]

## Leadership and team

As of the Series B in March 2026, Sunday had grown to roughly **70 engineers and researchers**, up from about 25 at the time of launch. The team is heavy on alumni of Tesla (Autopilot, Optimus, AI Infrastructure), [Google DeepMind](/wiki/google_deepmind), [Waymo](/wiki/waymo), [Meta](/wiki/meta_platforms), [OpenAI](/wiki/openai), [Apple](/wiki/apple_inc), and [Neuralink](/wiki/neuralink). In November 2025, [Electrek](/wiki/electrek) reported that Sunday had recruited "a full stack" of AI and robotics engineers away from Tesla. In addition to Zhao and Chi, public reporting has identified Perry Jia (data operations, ex-Tesla Optimus and Autopilot), Nishant Desai (ML, ex-Tesla Autopilot), Nadeesha Amarasinghe (AI infrastructure, ex-Tesla), and Kyle Miller (a BattleBots alumnus) as members of the founding team.[7]

## Reception and criticism

### How was Memo received?

Reception in late 2025 and early 2026 was unusually positive for a humanoid-robot startup, partly because Sunday chose to lead with multi-minute, edit-light demo videos rather than the heavily cut clips typical in the category. Outlets including SiliconANGLE, eWeek, Interesting Engineering, and Yahoo Finance highlighted Memo's ability to handle long-horizon tasks (most notably the table-to-dishwasher demonstration) in homes the robot had never seen.[1][8][9]

### What are the main criticisms?

Industry coverage also carries a thick layer of skepticism. Household humanoid robotics is a graveyard of well-funded predecessors: Jibo, [Anki](/wiki/anki), and Mayfield Robotics all raised real money and shipped real product before collapsing. Sunday will have to manufacture Memo at scale, hit something close to its sub-$10,000 target, and avoid the reliability problems that have historically tripped up consumer robotics. The Series B effectively gives the company about 18 months to prove out manufacturing and beta deployment.[10]

There is also debate about the data approach. Critics of the glove model argue that 4-DOF grippers lack force-torque feedback in some configurations and may limit the long tail of tasks Memo can ever learn; supporters argue that Memo only needs to do the common chores well, and that the glove network can scale faster than any teleoperation farm.[12]

## See also

- [MEMO (robot)](/wiki/memo_robot)
- [Humanoid robot](/wiki/humanoid_robot)
- [Imitation learning](/wiki/imitation_learning)
- [Diffusion Policy](/wiki/diffusion_policy)
- [Action Chunking with Transformers](/wiki/action_chunking_with_transformers)
- [ALOHA](/wiki/aloha)
- [Universal Manipulation Interface](/wiki/universal_manipulation_interface)
- [Service robot](/wiki/service_robot)
- [Physical Intelligence (company)](/wiki/physical_intelligence_company)
- [Figure AI](/wiki/figure_ai)
- [1X Technologies](/wiki/1x_technologies)
- [Tesla Optimus](/wiki/tesla_optimus)

## References

1. Wheatley, Mike. "Sunday wants to put a robot in every home, beginning with the launch of Memo." SiliconANGLE, November 20, 2025.
2. "DeepMind, Tesla Vets Emerge From Stealth With 'Sunday Robotics,' Backed by Conviction." Humanoids Daily, November 2025.
3. Wiggers, Kyle. "Humanoid robotics maker Sunday reaches $1.15B valuation to build household robots." TechCrunch, March 12, 2026.
4. Zhao, Tony Z.; Kumar, Vikash; Levine, Sergey; Finn, Chelsea. "Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware" (ALOHA / ACT). Robotics: Science and Systems, 2023.
5. Chi, Cheng; Feng, Siyuan; Du, Yilun; Xu, Zhenjia; Cousineau, Eric; Burchfiel, Benjamin; Song, Shuran. "Diffusion Policy: Visuomotor Policy Learning via Action Diffusion." Robotics: Science and Systems, 2023 (arXiv:2303.04137).
6. "Home Robots Finally Arrive: Inside Sunday Robotics." Bain Capital Ventures, March 2026.
7. Lambert, Fred. "Tesla is bleeding AI talent to a small new robotics start-up." Electrek, November 24, 2025.
8. "Sunday Unveils 'Memo': A Wheeled, Domestic Robot That Learns From $200 Gloves." Humanoids Daily, November 2025.
9. "Household humanoid robot trained on 10 million chores unveiled in US." Interesting Engineering, November 2025.
10. "Household Humanoid Maker Sunday Hits $1.15B Valuation." There's a Robot For That, March 2026.
11. Sunday Robotics. "ACT-1: A Robot Foundation Model Trained on Zero Robot Data." sunday.ai journal, November 19, 2025.
12. Wiggers, Kyle (citing industry analysis). "Sunday emerges from stealth with $35M for household robot called Memo." The AI Insider, November 20, 2025.
13. "Sunday's Series B: No More Demos." sunday.ai journal, March 12, 2026.