By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
No PriorsNo Priors Ep. 141 | With Sunday Robotics Co-Founders Tony Zhao and Cheng Chi
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
At a glance
WHAT IT’S REALLY ABOUT
Sunday Robotics Builds Cheap, Safe Home Robots With Massive Real-World Data
- Sunday Robotics co-founders Tony Zhao and Cheng Chi describe how recent advances in imitation learning, diffusion policies, and low-cost hardware are finally enabling general-purpose home robots. They explain the shift from brittle, hand-engineered classical robotics to scalable, data-driven systems powered by millions of real-world trajectories collected via custom gloves and tools like Umi. The founders emphasize full-stack integration—hardware, data collection, training pipelines, and product design—to achieve dexterity, generalization, and safety at consumer price points. They outline a 2026 home beta program, discuss realistic timelines and costs, and explain how to critically interpret robotics demos in an industry full of hype.
IDEAS WORTH REMEMBERING
5 ideasData-driven imitation learning is finally making general-purpose home robots viable.
Diffusion policies and transformer-based models trained on paired observation–action data have overcome the brittleness of classical, hand-engineered pipelines, allowing robots to learn complex manipulation from demonstrations at scale.
Scalable, high-quality data collection is the core bottleneck in robotics progress.
Sunday’s custom glove and Umi gripper setups let hundreds of non-expert humans collect millions of diverse trajectories in real homes, turning three students’ efforts into one of the largest real-world robotics datasets.
Full-stack integration is a competitive advantage in robotics.
Building hardware, data collection tools, cleaning pipelines, and training recipes under one roof lets Sunday iterate rapidly, fix failure modes end-to-end, and avoid research directions that don’t scale to real products.
Robot hardware can be cheaper, softer, and less precise when paired with strong perception.
Because modern robots ‘have eyes,’ low-cost, compliant actuators can rely on AI to correct mechanical inaccuracies in real time, enabling safe, home-ready systems without industrial-grade precision motors.
Data quality and distribution matching matter as much as data volume.
Failures like a robot trained only on rainy data breaking under direct sunlight show that broad generalization requires careful coverage of real-world conditions plus automated monitoring, calibration, and filtering pipelines.
WORDS WORTH SAVING
5 quotesWe’re kind of in between the GPT moment and the ChatGPT moment for robotics.
— Tony Zhao
In order for the robot to work in a sunny environment, it must have seen sunny environments in the training data.
— Cheng Chi
We really think the robot should have a face. It should have a cute face, and it should be very friendly.
— Tony Zhao
Data quality really matters. I always knew it, but once you scale it up, it really matters.
— Cheng Chi
When we look at demos, only index on things that are shown, and that’s likely the full scope of that task.
— Tony Zhao
High quality AI-generated summary created from speaker-labeled transcript.
