Sunday Robotics Raises $165M, Hits $1.15B Valuation

The Data Problem That Sunday Solved

Home robotics has been stuck in a loop for years. Building a robot that can reliably handle the chaos of a real kitchen, with its mismatched plates, sticky counters, and wine glasses that shatter if you grip them wrong, requires enormous amounts of training data. But you can't collect that data without deploying robots at scale, and you can't deploy robots at scale without the intelligence that comes from having the data. Tesla spent a decade collecting driving data from millions of vehicles before its self-driving systems started working well. For home robots, no equivalent data pipeline existed.

Sunday's founders recognized this deadlock early. Their solution bypasses the robot entirely during data collection. Instead of expensive teleoperation rigs (which can cost tens of thousands of dollars per setup and require trained operators), Sunday built the Skill Capture Glove, a roughly $400 device that mirrors Memo's hand kinematics exactly. A distributed network of over 500 "Memory Developers" across U.S. households wears these gloves while doing normal chores. Every plate they pick up, every sock they fold, every glass they place in the dishwasher rack becomes training data.

Because the glove and Memo's hand share identical geometry and sensor layouts, the recorded movements transfer directly to the robot's learning pipeline with a 90% success rate. No simulation gap. No domain adaptation headaches. Sunday calls this process Skill Transform, and it converts raw glove data into robot-equivalent demonstrations that look as if the robot itself performed them.

Under the Hood: ACT-1 and Memo's Capabilities

Memo runs on ACT-1, Sunday's foundation model trained entirely on human demonstration data collected through the glove pipeline, with zero traditional robot teleoperation data. The model builds on research that both founders published during their time at Stanford. Tony Zhao's original ALOHA project introduced Action Chunking with Transformers, which predicts short sequences of robot actions rather than individual timesteps, making it far easier to learn complex manipulation behaviors from limited demonstrations. Cheng Chi's Diffusion Policy work showed that diffusion-based models handle the messy, multi-modal nature of real-world manipulation better than prior approaches.

The result is a system capable of what Sunday calls "ultra long-horizon mobile manipulation." Their flagship demo, Table-to-Dishwasher, involves Memo autonomously clearing a dinner table of plates, utensils, and wine glasses, dumping food waste, loading items into the dishwasher, and starting the cycle. That single task chain includes 33 unique dexterous interactions with 21 different objects across more than 130 feet of navigation. The system generates actions at varying granularity, from millimeter-precision glass placement to meter-scale room navigation, all in one continuous run.

AI-generated image

Memo's vertically integrated hardware stack lets Sunday iterate on mechanical design and AI simultaneously.

Memo itself is a wheeled platform (not bipedal) with dual articulated arms. Sunday chose wheels over legs for a pragmatic reason: homes have flat floors, and wheels are quieter, more energy-efficient, and far simpler to engineer reliably than bipedal locomotion. The robot's hands are custom-designed to balance human-like dexterity with manufacturing constraints, optimizing for the grip patterns that matter most in household tasks. Force sensitivity is fine enough that ACT-1 can hold two wine glasses in a single hand without breaking them.

How Sunday Compares to Other Home Robot Efforts

The Founders and Their Research Roots

Tony Zhao (CEO) and Cheng Chi (CTO) met on X after independently publishing papers that reshaped how the robotics field thinks about learning from demonstrations. Tony's ALOHA project at Stanford introduced a low-cost teleoperation system paired with Action Chunking with Transformers (ACT), a method that made it possible to learn difficult manipulation behaviors from just minutes of demonstration. Cheng's work at Columbia on Diffusion Policy proved that diffusion models could generate robot action trajectories more reliably than earlier techniques, especially for tasks with complex, multi-step motions.

They started Sunday in April 2024, working out of a garage in Mountain View. Within 18 months, they emerged from stealth with working demos that caught Wired's attention and landed them on the radar of top-tier investors. The company's approach, solving the data bottleneck through clever hardware design rather than brute-force teleoperation, reflected a pattern visible across both founders' academic work: build cheap, practical systems that scale.

What This Means for Household Robotics

Sunday's funding round arrives during a period of massive capital inflow into humanoid and household robotics. In February 2026 alone, Apptronik raised $520 million backed by Google and Mercedes-Benz, while Mind Robotics (the Rivian spinout) closed a $500 million Series A this week. But those companies are targeting factories and warehouses. Sunday is one of very few well-funded startups going directly after the consumer household market.

The household segment is notoriously difficult. Homes are unstructured environments where layouts change constantly, objects vary wildly, and robots need to operate safely around children, pets, and breakable items. Previous attempts at household robots (think Jibo, Kuri, and even early Roomba competitors) either failed commercially or settled for extremely narrow tasks. Sunday's bet is that their data flywheel, scaling from 500 Memory Developers today with plans to 5x that number this year, can accumulate enough real-world household data to handle the "long tail" of domestic chaos.

AI-generated image

Sunday's target environment: real American homes with all their unpredictable layouts and clutter.

The business model also looks different from industrial robotics. While companies like Figure AI sell or lease robots to corporate customers with predictable order volumes, Sunday needs to crack consumer pricing, marketing, and support at scale. No consumer price has been announced, but the economics of the Skill Capture Glove (~$400 per unit) and the distributed data collection model suggest Sunday is thinking about cost efficiency from the ground up. Bain Capital Ventures noted that Sunday deliberately recruits Memory Developers from U.S. households because those environments will match Memo's actual deployment conditions.

Frequently Asked Questions

What to Watch Next

Sunday's Thanksgiving 2026 deadline is the number to circle on the calendar. That is when the company plans to put its first autonomous Memo units into real homes. If the beta launch goes well, expect a rapid scaling announcement in early 2027, likely paired with another funding round to finance consumer manufacturing. The company's plan to 5x its Memory Developer network this year will be a leading indicator: more data collectors means more skills, which means a more capable robot at launch.

The competitive picture is also shifting fast. While Sunday targets households, industrial players like Figure AI, Apptronik, and Tesla are accumulating their own fleet deployment data from factory environments. If any of those companies pivots toward consumer markets (Tesla has hinted at this repeatedly), Sunday's head start in household-specific data could prove decisive, or it could face well-funded competition from multiple directions.

The home robot dream has burned through billions of dollars in venture capital over the past two decades without producing a single commercially successful product. Sunday's founders believe the missing piece was always data, not hardware. With $165 million in fresh funding and a growing army of glove-wearing data collectors spread across American kitchens, they are about to find out if they are right.