Sunday Robotics raises $165M Series B at $1.15B valuation, shifts focus from demos to real home deployments
Mar 16, 2026 with Tony Zhao
Key Points
- Sunday Robotics raises $165M Series B at $1.15B valuation to fund real home deployments this year, abandoning the demo-heavy approach that has defined robotics startups.
- The company collects training data through gloves that mirror robot hand movements, letting thousands of people gather high-precision tactile data across diverse home environments.
- CEO Tony Zhao argues synthetic data lifts everyone but real-world fidelity still closes the final gap from 95% to 99.99% accuracy in manipulation tasks.
Summary
Sunday Robotics raised $165 million in Series B funding at a $1.15 billion valuation, led by Kotu. CEO Tony Zhao announced the company is moving away from demos and shifting its entire focus to real-world home deployments this year.
Zhao frames the home as the long-term proving ground for physical intelligence because of its inherent complexity. The initial beta program will target high-friction household chores: laundry, dishes, organization, and cleaning. These are the tasks people spend the most time on and hate the most.
Data collection
Sunday's proprietary advantage sits in how it collects training data. Rather than deploying thousands of robots, the company manufactures gloves that mirror the robot's hand movements. Users wear the gloves in their own homes and collect data during everyday activities. This approach yields high-quality hand data with precise movements, force feedback, and tactile information. It also yields high diversity across thousands of people performing tasks in varied environments, and high volume since people can wear gloves part-time on flexible schedules.
Zhao plans to scale to thousands of data collectors this year to build a proprietary dataset for training the foundational physical intelligence model. Egocentric video data from sources like GoPro footage lacks precision and loses force information. Sunday's strategy is to blend public internet video with proprietary glove data. Public data extracts broad knowledge about physics and how spaces are arranged. Proprietary data bridges the gap from general knowledge to a deployable, valuable product.
Synthetic data and world models
Zhao sees synthetic data generation from world models as valuable for improving all models. But he is skeptical it will close the final gap. Jumping from 95% to 99.99% accuracy, or solving the last millimeter of a manipulation task, still requires real data fidelity. Synthetic data helps with breadth. Real data is needed for precision.
Home is the long-term goal because it is messy enough to drive genuine physical intelligence. As the robot's capabilities grow, offices, hotels, and other structured environments will unlock as secondary opportunities. Zhao is keeping that door open but has not prioritized it.
The bet is concrete: deployments in real homes this year, not more polished demos.