No Priors Podcast: Sunday Robotics — Scaling the Home Robot Revolution

Episode: Sunday Robotics: Scaling the Home Robot Revolution with Co-Founders Tony Zhao and Cheng Chi
Date: November 19, 2025
Hosts: Elad Gil, Sarah Guo
Guests: Tony Zhao and Cheng Chi, Co-Founders of Sundae (makers of Memo, the first general home robot)

Episode Overview

This episode dives deep into the state and future of home robotics, as seen through the lens of Sundae’s co-founders Tony Zhao and Cheng Chi. The conversation tracks the trajectory from foundational AI and imitation learning research to building Memo, the company’s first home robot. Key topics include breakthroughs in scalable data collection, the unique challenges in robotics hardware and software, design philosophy for consumer robots, and pragmatic reflections on timelines and market readiness.

State of AI & Robotics: Why Now?

[00:55–03:13]

• Inflection Point in Robotics:

  • Robotics is at the "in-between" stage: there’s a proven recipe from foundational AI, but the field hasn't yet delivered mass-market, reliable consumer robots.
  • Tony Zhao: “I think we’re kind of in between the GPT moment and the ChatGPT moment... we have a recipe that can be scaled, but we haven’t scaled it up yet. And...we haven’t scaled up so much so that we can have a great consumer product out of it.” [01:15]

• Classical vs. Modern Approaches:

  • The old approach (sense, plan, act) required massive human intervention for every new task/environment—resulting in slow, non-generalizable progress.
  • Cheng Chi: “For every task that means a paper...you throw away all your code, all your work, and you start over again. And that’s also…what happened to industry.” [02:24]

Major Research Breakthroughs and Methods

[03:13–11:30]

• Diffusion Policy for Imitation Learning:

  • Enables capturing diverse behaviors from multiple people—not just teleop experts—expanding scalable data collection.
  • Cheng Chi: “Diffusion model really allows us to capture multiple modes of behavior for the same observation in a way that’s still preserved training stability.” [03:23]

• Aloha System & ACT Paper:

  • Made data collection much more intuitive, “like playing a video game” [05:14], and allowed for deeply dexterous data with low lag.

• Transformers for Robotics:

  • Transformers (borrowed from LLMs) only worked once high-quality, dexterous data and trajectory-level predictions (“action chunking”) were available.
  • Tony Zhao: “Once you have very strong and dexterous data sets, just throw a transformer at it and it works quite well actually.” [06:05]

• Innovation in Data Collection — Umi Glove:

  • A glove interface with GoPro tracking enabled large-scale, diverse data gathering outside the lab.
  • Cheng Chi: “We just took the grippers everywhere...very quickly we got...1,500 video clips...and that turns out to be one of the biggest data sets in robotics...by three people.” [08:05]

From Research to Company: Building Sundae

[10:38–13:07]

• Team Expansion and Vision:

  • From hacking in an apartment to growing a full-stack team of 30–40 people encompassing mechanical engineering, supply chain, software, and AI.
  • Mission: “To put a home robot in everyone’s home.” [11:30]

• Design Philosophy:

  • Prioritize simplicity and scalability: the home robot has a cute, cartoon-inspired look, is friendly, and is intentionally NOT humanoid or over-complex.
  • Tony Zhao: "Whenever we see something that we can accelerate it with simplification, we'll go simplify that." [13:07]

Hardware Innovations and Tradeoffs

[13:07–15:05]

• From Stiff Industrial Arms to Compliant, Perceptive Robots:
  • Old robots were fast and blind. Now, AI vision allows for cheaper, compliant (“soft”) actuators.
  • Cheng Chi: “Because of the breakthrough we had in AI, now robot have eyes so it can actually correct its own mechanical and hardware inaccuracies...that opened up a new...design space.” [14:24]

Scalability, Data, and Full-Stack Challenges

[15:05–21:49]

• Scaling Up: Challenges and Learnings:

  • Beta program launching in 2026, with internal prototypes being refined for in-home trials.
  • Surprises: Generalization/dexterity improved far beyond expectations with data scale; scaling up was harder and more “painful” (engineering, reliability, hardware quality) than anticipated.
  • Tony Zhao: “I think at the beginning...if we build this, someone can just take our glove and they’ll build the same thing...but as we go along...it turns out things are so much harder than we thought it was.” [17:24]

• Quantifying Progress:

  • Sundae has collected nearly 10 million “trajectories” (complex activity sequences) in the wild—an industry-leading scale.
  • Tony Zhao: “At this point, we are almost 10 million trajectories...those trajectories are not just like, oh, pick up a cup. It's these long...tasks.” [19:16]

Data Collection Modalities: Teleop, Glove, RL, Simulation

[19:50–24:55]

• Teleop vs. Glove:

  • Initially worried glove-collected data would be lower quality; in practice, it encourages more dexterous, natural movements and matches or exceeds teleop for usefulness.
  • Requires sophisticated data conversion (“engineering series”) to map human demonstrations to robot actions.

• Role of RL (Reinforcement Learning):

  • Good for locomotion (simulating rigid body dynamics); sample-inefficient and less suited to manipulation (hard to simulate real-world variety).
  • Tony Zhao: “If I have a perfect world simulator, anything can be done there...but for robotics...some things are harder than others.” [21:49]

• Data Quality at Scale:

  • As data is scaled up, managing diversity and quality is harder and ever more important; need for robust automation in calibration, validation, and failure detection.

Full-Stack Approach & Technical Challenges Ahead

[25:11–27:59]

• Research at Scale:
  • Now possible to focus on scaling robustness — “figuring out the training recipe at scale.” [25:11]
• Hardware Remains Hard:
  • Reliability and performance envelopes are still unknown; mechanical and learning teams iterate rapidly after device failures.
  • Building everything in-house (hardware, software, data pipeline) considered essential due to the high standard and rapidly changing “definition of good."
  • Tony Zhao: “Before we start to...dive into research, we actually focus on the infrastructure and...scalable data pipeline..." [26:22]

Timeline, Costs, and the Future Home

[27:59–30:47]

• When will home robots reach consumers?

  • Beta program in 2026; mass-market timeline depends on beta learnings but is “definitely not a decade away.” [29:08]

• Current and Projected Costs:

  • Prototype robots cost $6,000–$20,000 (mainly from small-scale parts and custom cladding).
  • With scale (thousands of units), cost “likely under 10k"; launch price aligned accordingly.

• Life with Home Robots:

  • Tony Zhao: “The world we’ll live in...the marginal cost of labor in homes goes to zero.” [30:47]
  • Cleaner homes, less time on chores, more time for hobbies, family.

Evaluating Robotics Demos (and Hype)

[31:02–36:54]

• How to Interpret Demos:

  • Be skeptical: check for autonomy, task diversity, and generalization, not just one-off tricks for the camera.
  • Tony Zhao: “If you see a robot handing one drink to one person, first ask...is that autonomous or is that teleoperated?...does it show giving another slightly different color cup to the same person or not?” [31:02]

• What Memo Can Actually Do:

  • Demos showcase:
    • Long-horizon tasks like clearing a table, loading dishwashers
    • Generalization: zero-shot performance in Airbnbs using only prior data
    • Dexterity: folding socks, operating espresso machines, handling fragile objects
  • Glove-collected data allows scaling of long, multi-step, nuanced behaviors that are generalizable and reliable.

Building the Team and Company Philosophy

[36:54–38:47]

• Who Sundae is Hiring:

  • Full-stack roboticists—people who want to learn mechanical, electrical, software, and ML together.
  • Cheng Chi: "We have a couple examples of training just full time software engineers to become robotic training engineers to become roboticists. And so if you want to learn about robotics...let us know.” [37:45]

• Learning and Growth:

  • Personal journeys from hardware design to programming to ML show the broad, multidisciplinary scope of robotics work.
  • Tony Zhao: “We're just fundamentally such a full stack company, we're not just about the software, we're not just about the hardware, but...the whole experience, the whole product, and making sure...is general and scalable in the future.” [38:25]

Notable Quotes

• On the Home Robot Market:
  • Tony Zhao: “Nobody want to do their dishes, nobody want to do their laundry. People will love to spend more time with their family, with their loved ones...if the robot is cheap, safe and capable, everyone will want our robot.” [00:05]
• On Scaling Robotics Data:
  • Sarah Guo: "You spent like...$200,000 across all of your academic research. And yet the scale of data collection has translated to model capability is leading." [09:39]
• On Hardware Complexity:
  • Cheng Chi: “Hardware is hard, but it is important and I think it’s a hard but right thing to do. And I think we as a field shouldn’t avoid doing the hard things just because they’re hard.” [26:07]
• On Future Impact:
  • Tony Zhao: “I think the world we’ll live in is…it’s going to be cleaner. And...the marginal cost of labor in homes goes to zero.” [30:47]

Key Timestamps & Topics

• [00:55] — Context: Why robotics is exciting right now
• [03:23] — Diffusion policy and scalable imitation learning
• [07:06] — Umi glove innovation and its impact
• [10:38] — Founding Sundae, building the team and vision
• [13:07] — Home robot design—why not just mimic humans?
• [14:24] — Actuators, compliance, and safety in home robots
• [17:24] — Scaling up: surprises and pain points
• [19:16] — Data scale: Sundae’s 10 million "trajectories"
• [21:49] — RL vs. imitation learning for robotics
• [25:11] — Technical obstacles: training recipes and hardware
• [27:59] — When will robots reach homes? Beta timeline and cost insights
• [31:02] — Evaluating demos and robotics hype
• [32:36] — Sundae demos: real generalization and dexterity
• [36:54] — Building the team: full-stack needs and multidisciplinary learning

Tone and Closing Reflections

Throughout, the conversation blends grounded optimism with candid realism. Zhao and Chi emphasize relentless iteration, humility before hardware’s complexity, and the belief that general, affordable robotics is both technically near and societally transformative.

Summary prepared for readers who have not listened—skip to any timestamp above to dive deeper into the world Sundae is building, and how the future of home robotics is closer than you might think.