Sunday Hit $1.15B Valuation Building a Robot for Chores

In April 2024, two robotics researchers moved into a Mountain View apartment and started building a home robot from scratch. Twenty-three months later, Sunday closed a $165M Series B at a $1.15B valuation, becoming one of the fastest robotics startups ever to reach unicorn status. The robot they built, Memo, can fold your socks, load your dishwasher, and pull an espresso shot without ever having trained on robot data.

Sunday's core bet is that the best way to teach a robot to live in a home is to watch humans live in their homes. The company deployed over 2,000 motion-capture gloves into real US households and recorded more than 10 million activity episodes. That data trained ACT-1, Sunday's foundation model, which runs entirely onboard Memo's wheeled torso. Sunday is targeting 50 beta households by mid-2026 and a full consumer launch by Thanksgiving 2026.

AI-generated image

Memo's four-fingered silicone hands are IP67-rated and modeled after the gloves used to capture human training data, closing the loop between data collection and deployment.

The Sunday Story

Tony Zhao and Cheng Chi are not typical startup founders. Zhao earned his EECS degree from UC Berkeley in 2021, then enrolled in Stanford's CS PhD program, advised by Sergey Levine and Chelsea Finn, two of the most cited researchers in robot learning. He had already shipped real work: the ALOHA project at Berkeley demonstrated low-cost robotic imitation learning that caught the attention of Google DeepMind, where he later worked. He also spent time on Tesla Autopilot. In early 2024, he dropped out of Stanford to start Sunday.

Chi's background is equally concrete. He completed his PhD in computer science at Columbia University under Shuran Song, where he invented Diffusion Policy, a robot learning algorithm that has since become one of the most widely adopted methods in the field. He collected two Best Paper Awards and two Best Systems Paper Finalist recognitions during his PhD. He holds a BS in CS from the University of Michigan and serves as Sunday's CTO.

They founded Sunday in April 2024 and started building in their apartment. No lab, no university affiliation, no large team. Just two people and a hardware prototype. By December 2024, their single-armed early Memo could arrange shoes, the first task it learned. By February 2025, they had moved into a 10-person hacker house in Mountain View and expanded the team. By October 2025, Memo had added sock folding, glassware handling, and espresso to its repertoire.

Sunday emerged from stealth on November 19, 2025. The announcement included Memo's full specs, a live demo, the ACT-1 model reveal, and an open beta application. The waitlist hit 3,000 within days. Less than four months later, on March 12, 2026, Sunday announced a $165M Series B led by Coatue Management, with participation from Bain Capital Ventures, Fidelity, Tiger Global, and Benchmark. The round was oversubscribed. The valuation: $1.15 billion. Sunday's total funding reached approximately $200M, combining the Series B with a prior ~$35M seed round.

The team has grown to roughly 30 engineers and researchers as of March 2026. The pace from apartment prototype to an oversubscribed unicorn round is one of the faster timelines in robotics history, a field where hardware development cycles typically span years, not months.

Under the Hood: Memo's Architecture

Memo is not a bipedal humanoid. That choice is intentional. Sunday designed Memo as a wheeled-base torso humanoid, a platform optimized for the real constraints of household work. Homes have flat floors, doorways, counters, and cabinets. You do not need legs to navigate them. You need reach, dexterity, and reliability. Memo's wheeled base keeps the center of mass low, eliminates fall risk entirely (the robot maintains passive stability even when powered off), and delivers a 1 m/s top speed that is practical on hardwood, tile, and carpet.

Physical Specs

AI-generated image

Memo's wheeled base and adjustable torso make it well-suited for kitchen tasks. Countertop clearance, overhead cabinets, and floor-level appliances are all within reach without requiring legs.

ACT-1: The Foundation Model

ACT-1 is Sunday's core differentiator. The model was trained exclusively on human activity data: no robot demonstrations, no simulated environments, no synthetic data. The training corpus consists of 10 million+ episodes captured by the Skill Capture Gloves deployed in 2,000+ real US homes. Each glove records motion, force, and visual data from human hands performing household tasks: loading a dishwasher, folding a shirt, picking up a fragile glass.

The result is a model that generalizes to unseen homes. ACT-1 builds a 3D map of a new environment and adapts its task plans to the specific layout. Sunday's November 2025 demo included a table-clearing sequence that required 33 discrete object interactions, coordination with 21 distinct objects, and navigation totaling 130 feet through a home. The model completed it without human intervention. Sunday describes the system as privacy-focused: local processing, no cloud video stream, maps not retained after task completion.

The connection between the Skill Capture Gloves and Memo's hands is not accidental. The glove geometry was designed to match Memo's four-finger gripper profile, so the kinematic data transfers directly without remapping. This systems-level thinking reflects founders who have shipped real robotics research. Zhao's ALOHA work involved exactly this kind of end-to-end pipeline design, and it shows in how Sunday structured ACT-1's data pipeline from day one.

How Memo Compares

The cost gap is the most striking number in that table. Memo's ~$20K build cost sits 5x to 12x below the nearest industrial competitor, and Sunday's target of sub-$10K retail is in a category of its own. The wheeled architecture contributes directly: bipedal locomotion systems add actuators, sensors, and control complexity that drives cost up sharply. By skipping legs, Sunday cut the bill of materials and simplified control while keeping the arm dexterity needed for household work.

Sunday: The Full Picture

Founders and Team

Sunday's founding team carries unusually deep credentials for a company less than two years old. Tony Zhao (CEO) published on robotic imitation learning at a level that drew Stanford PhD admission and Google DeepMind interest before he turned 25. The ALOHA project demonstrated that low-cost teleoperation hardware could generate enough data to train capable manipulation policies, and it is one of the most-cited recent papers in the field. That exact insight underpins Sunday's Skill Capture Glove program.

Cheng Chi (CTO) invented Diffusion Policy during his Columbia PhD, a method now used by robotics labs worldwide for training manipulation skills. Two Best Paper Awards across top robotics venues is a rare distinction. Chi's depth in generative modeling and policy learning is directly reflected in ACT-1's architecture and training approach. His advisor, Shuran Song, is one of the leading researchers in robotic manipulation, and the lineage shows in how Sunday approaches the data-to-policy pipeline.

The broader team of 30+ includes engineers and researchers from robotics, machine learning, and hardware backgrounds. A team that ships a beta-ready household robot in under two years, from an apartment starting point, is not doing so by hiring generalists.

Funding Timeline

Coatue is a known growth investor in deep tech. Benchmark's participation stands out: the firm's portfolio includes Snap, Uber, and other high-velocity consumer businesses, and its inclusion signals conviction that Memo is a mass-market consumer opportunity, not just a research curiosity. An oversubscribed round at these terms in the current rate environment carries meaningful signal.

Key Milestones

AI-generated image

Sunday's target: a robot that operates in the background while families go about their lives. Memo runs at roughly 50% of human pace for safety, keeping it unobtrusive in occupied spaces.

What Memo Means for Home Robotics

The home robotics category has a long history of overpromising. Roomba solved vacuuming in 2002. Everything else, laundry, dishes, cooking, has remained stubbornly human. The gap between a robotic arm that picks up a specific part in a controlled factory and one that picks up any of 21 objects on a real kitchen table is enormous. Sunday is the first company with a credible technical path to closing that gap at consumer pricing.

The key factor is the data flywheel. Sunday's 2,000+ Skill Capture Gloves are not a one-time data collection event; they are an ongoing pipeline. As the beta program expands to 50 households, the company accumulates real-world task data at a rate no competitor with a lab-only training approach can match. Each Founding Family home adds to the distribution of environments, objects, and edge cases that ACT-1 learns from. The model improves, new tasks become learnable, and the gap between Memo and a conventional robot widens with each month of deployment.

Pricing is the other structural advantage. Industrial humanoid robots like Digit (~$250K+) and Figure 02 (~$100K+) are priced for enterprise buyers with capital budgets and ROI models. A $10K target retail price puts Memo in a different conversation entirely. At that price point, Memo competes with high-end appliances, not factory automation systems. The buyer is a homeowner, not a procurement manager.

Timing matters too. Sunday is targeting a Thanksgiving 2026 consumer launch. Figure AI's Figure 02 is deployed at BMW manufacturing plants. Agility Robotics' Digit is running at Amazon warehouses. Tesla Optimus Gen 2 is in internal Tesla deployment. None of the major funded humanoid programs are shipping to households in 2026. Sunday has the home market to itself for the near term, and it is using that window to build deployment data that competitors will not have.

The 12-Month Outlook

With $165M in fresh capital and a $1.15B valuation, Sunday's next 12 months will determine whether Memo transitions from compelling demo to real product. The company has laid out a specific roadmap: expand the Founding Families beta to 50 households, collect real-world deployment data at scale, and ship consumer units by Thanksgiving 2026.

The beta program is the most important datapoint to watch. Fifty real households means 50 different kitchen layouts, appliance configurations, object sets, and user behaviors. ACT-1's ability to generalize across all of them, without per-home customization beyond the 3D mapping step, is the central technical claim Sunday needs to prove. Consistent performance on core tasks during beta makes the consumer launch credible. Systematic failure modes would need to be addressed before a public rollout, and Sunday has capital and time to do that if needed.

The $165M Series B will fund three things: hardware manufacturing scale (to bring build cost toward the sub-$10K retail target), ACT-1 model expansion (new skills, improved generalization, broader task coverage), and team growth. Based on typical post-Series B headcount trajectories in robotics, the team of 30+ is likely to double within 12 months.

Competitive pressure will intensify. Boston Dynamics launched its Atlas commercial program in 2025. Figure AI is generating revenue at BMW. Agility Robotics has Amazon as a reference customer. Each of those programs generates real-world deployment learnings that could eventually translate toward home applications. Sunday's head start in home-specific data and its pricing advantage are real, but they are not permanent moats in isolation.

The other variable is task expansion. Memo currently handles six demonstrated task categories. Each new skill requires glove-captured training data and model updates. Sunday's 2,000+ deployed gloves give it a scalable path to adding skills without new hardware, but pace matters: how many skills can ACT-1 absorb per quarter, and which ones will drive consumer purchase decisions at the $10K price point?

Watch for: beta household count updates, ACT-1 skill announcements, team hiring velocity, and any manufacturing partnership announcements ahead of the consumer launch hardware run. If Memo ships to Founding Families by mid-2026 and performs consistently, the Thanksgiving consumer launch becomes very real.

Frequently Asked Questions