Tony
Zhao

The PhD That Built a Unicorn from the Scrap Pile

In 2023, a third-year Stanford PhD student named Tony Zhao posted a paper about a bimanual robot that could thread a zip tie, juggle a ping pong ball, and assemble a chain - built for roughly $20,000 in parts. He called it ALOHA. He published everything. The code, the designs, the data collection approach. Free. Open. His advisor was Chelsea Finn, one of the leading voices in robot learning research. Nobody expected the paper to become a company valued at over a billion dollars in under two years.

But that is exactly what happened.

Zhao co-founded Sunday in 2024 alongside Cheng Chi, his fellow Stanford PhD researcher. The name evokes leisure - the day you are supposed to rest. The robots are being built so that every day can feel more like one. The mission is specific and unhyped: put a helpful robot in every home. Not to do a party trick. Not to wander around looking humanoid. To handle the dishes, the laundry, the tidying - the relentless background hum of domestic labor that never actually ends.

Before Sunday, Zhao moved through the elite corridors of AI research like someone who already knew he was building toward something specific. A UC Berkeley EECS graduate (class of 2021, advised by Sergey Levine and Dan Klein), he joined Stanford's CS PhD program and picked up the Stanford Robotics Fellowship. Along the way he spent time at DeepMind as a part-time researcher and interned at Tesla Autopilot and Google X Intrinsic. He was accumulating leverage, not a resume.

The insight he was chasing: robots don't need to be expensive or exotic to be useful. They need to be teachable. The problem was data. How do you get a robot to learn from millions of real household demonstrations without it taking decades? Zhao's answer came in layers - first ALOHA, then ACT, then Mobile ALOHA - each paper tightening the loop between human movement and machine replication.

ALOHA, ACT, and the Art of Teaching Robots

The ALOHA papers are not obscure academic footnotes. They circulated through the robotics community like samizdat - shared, forked, built upon. Zhao's core contribution was deceptively simple: stop trying to hard-code robot behavior and start showing the robot what to do, over and over, until it gets it. Imitation learning at scale.

• 01
  ALOHA - A Low-cost Open-source Hardware System for Bimanual Teleoperation

  A $20,000 dual-arm system for precise manipulation. Threading zip ties. Juggling ping pong balls. Open-sourced entirely - hardware, software, datasets.
• 02
  ACT - Action Chunking with Transformers

  The algorithm behind ALOHA's learning. Instead of predicting one action at a time, ACT predicts in "chunks" - reducing compounding error and making complex manipulation tractable.
• 03
  Mobile ALOHA

  Mounted on a mobile base, the system demonstrated sauteing shrimp, cleaning wine spills off floors, calling elevators, and giving high-fives. It went viral for the shrimp alone.
• 04
  ALOHA 2 & ALOHA Unleashed

  Advanced iterations pushing the hardware and policy limits further, co-authored with researchers across Google DeepMind and Stanford.

Memo: No Legs, No Problem

Launched in November 2025, Sunday's robot is named Memo. It rolls on a wheeled base. This is a deliberate choice in a market saturated with humanoid announcements - two-legged robots that stumble through demos and require years of perfecting balance before they can carry a plate. Zhao looked at the landscape and concluded that legs are a constraint, not a feature, for household work.

Memo does not walk. It navigates. And it cleans plates, loads dishwashers, strips laundry, makes espresso, and prepares simple dishes. Its intelligence comes from approximately 10 million episodes of real household task demonstrations collected through Sunday's custom glove-based teleoperation system - not simulation, not synthetic data, but actual human hands doing actual chores, captured at scale and used to train the robot's policies.

The Founding Family Beta opened alongside the November 2025 launch. Fifty households were selected to receive individually numbered Memo robots in late 2026, with direct support from the Sunday team. It is a small number by design - the kind of constrained rollout that prioritizes learning over metrics.

$200M and a $1.15B Unicorn in Under Two Years

In November 2025, the same month Memo launched, Sunday announced a $35M Series A led by Benchmark and Conviction. Less than four months later, in March 2026, the company raised a $165M Series B led by Coatue - with Thomas Laffont joining the board. Bain Capital Ventures, Fidelity, Tiger Global, Benchmark, Conviction, and Xtal Ventures all participated.

The speed is notable. Tesla reportedly lost AI talent to Sunday - Electrek documented the brain drain from Elon Musk's robotics ambitions to Zhao's smaller, faster-moving team. The Sunday roster includes engineers and researchers from Stanford, Tesla, DeepMind, Waymo, Meta, OpenAI, and Apple. Over 130 people, as of early 2026.

From Berkeley Lab to $1B Company

Tony Zhao on Scaling the Home Robot Revolution

In No Priors Episode 141, Zhao and co-founder Cheng Chi discuss what a GPT moment for physical AI actually looks like, how Sunday broke the data bottleneck, and why the shift from specialized to general-purpose robots is the defining bet of the decade.

No Priors Ep. 141 | With Sunday Robotics Co-Founders Tony Zhao and Cheng Chi

The Contrarian Bets Behind Sunday

The home robotics space is crowded with announcements. Zhao made a series of choices that look obvious in retrospect but were not obvious when he made them.

He chose wheels over legs. When Figure, Agility, and a dozen others were announcing humanoids, Zhao concluded that bipedal locomotion is an engineering tax for home use. Most household tasks happen at counter height. You don't need a gait. You need dexterity, reliability, and the ability to fit through a doorway.

He published before he commercialized. ALOHA and ACT were fully open-source. That seeded the field with his methodology and attracted the exact caliber of researchers who would later join his team. Generosity as a recruiting strategy.

He bet on real-world data over simulation. Many robot companies train in simulation and transfer to the real world, accepting the "sim-to-real gap" as a necessary evil. Zhao built a teleoperation system to collect 10 million episodes of actual household task demonstrations. The gap doesn't exist if you never left the real world.

He started small. Fifty Founding Family Beta households. Not fifty thousand. The deliberate constraint allows Sunday to instrument each deployment carefully, learn in the real environment where the robot will actually live, and iterate before scaling. It is the opposite of a press-release launch.

Zhao's academic background shows in how Sunday talks about the problem. The company does not promise the robot will do everything. It promises that Memo will handle the chores that happen repeatedly, predictably, in the same kitchen, in the same house. The learning is household-specific. The robot adapts. Over time, Memo builds a model of your home, your dishes, your routines. It is not trying to be a general-purpose machine on day one. It is trying to be useful on day one and get better from there.

That distinction - useful now versus impressive eventually - may be what separates Sunday from the field.