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

"We believe that if the robot is cheap, safe, and capable, everyone will want our robots." This statement from Tony Zhao and Cheng Chi, co-founders of Sunday Robotics, encapsulates the ambitious vision driving their work on Memo, the first general-intelligence personal robot. Sarah Guo, host of the No Priors podcast, sat down with Zhao and Chi to discuss the state of AI robotics and the path toward widespread home robot adoption.

The conversation highlighted the significant shift in the robotics industry, moving from specialized, task-specific machines to more adaptable, general-purpose robots. This evolution is largely catalyzed by recent advancements in AI, particularly in areas like diffusion policy and imitation learning. These breakthroughs have enabled robots to learn complex tasks through observation and to generalize that learning to new situations, a significant leap from previous methods.

The co-founders elaborated on the challenges they encountered while developing Memo. One key hurdle was the sheer scale of data collection required for real-world robot training. Traditional methods of data collection were proving too slow and expensive to achieve the necessary breadth and depth. To overcome this, Sunday Robotics developed an innovative glove system. This system allows human users to teleoperate the robot's actions, effectively collecting real-world data at scale and with high fidelity. "The glove system allows us to scale real-world data collection," explained Zhao, underscoring its importance in training Memo for diverse tasks.

The impact of diffusion policy and imitation learning was a recurring theme. These techniques allow robots to learn from demonstrations, mimicking human actions to perform tasks. This is a significant departure from older methods that relied on explicit programming for every action. "Imitation learning, enter UMI," as the discussion framed it, signifies a move towards more intuitive and human-like robotic capabilities. This approach not only speeds up the learning process but also enables robots to perform tasks that are difficult to explicitly program.

Sunday Robotics' design philosophy for Memo emphasizes a balance between capability, safety, and affordability. Chi elaborated on this, stating, "We believe that if the robot is cheap, safe, and capable, everyone will want our robots." This philosophy guides every aspect of Memo's development, from its hardware design to its software capabilities. The robot is envisioned as a personal assistant, capable of taking on household chores and freeing up human time for more meaningful activities.

The company is targeting a 2026 in-home beta program, with mass deployment projected to be only a handful of years away. This ambitious timeline is supported by the rapid progress in AI and the company's innovative approach to data collection and learning. They are currently hiring for various roles, indicating a strong push towards realizing their vision.

A crucial insight that emerged from the discussion is the distinction between past robotic demonstrations and the current state of the art. Many past demonstrations, while impressive, were often heavily staged and lacked the robustness needed for real-world deployment. Sunday Robotics aims to overcome this by focusing on robust, generalizable skills learned through real-world data and advanced AI techniques. The success of Memo, and the broader field of personal robotics, hinges on this ability to bridge the gap between controlled demonstrations and unpredictable, everyday environments.

The conversation also touched upon the importance of data quality at scale. While large datasets are crucial, ensuring the quality and relevance of that data is paramount for effective robot learning. Sunday Robotics' approach, leveraging human demonstration through their glove system, aims to capture high-quality, task-specific data that can be efficiently used to train their AI models. This focus on data quality, combined with advancements in AI, positions Sunday Robotics to potentially deliver on the long-held promise of truly capable and accessible home robots.