Sunday Robotics says its robot can fold clothes it has never seen in unfamiliar homes

In a significant leap forward for home robotics, Sunday Robotics has announced a breakthrough in their quest to create autonomous robots capable of performing household tasks in unpredictable environments. The Bay Area startup introduced ACT-2, an advanced AI model that powers Memo, their wheeled home robot, during a recent press event. What sets Memo apart is its remarkable ability to fold laundry successfully more than 99% of the time, even when faced with garments it has never encountered before. This achievement promises to reshape the landscape of home robotics, according to CEO Tony Zhao. "I anticipate that by 2026, we will see a dramatic reduction in the timeline for the deployment of autonomous home robots, thanks to this innovation," he stated. For years, developers have struggled to create robots that can adapt to new environments and tasks. Traditional robots often excel in controlled settings but falter when faced with unfamiliar objects or scenarios. The ACT-2 model, however, enables Memo to transfer knowledge gained in one setting to another, eliminating the need for extensive retraining in different homes. Zhao highlighted the importance of establishing new standards in robotics, proposing a framework called "Solve" to better evaluate a robot's capabilities. This initiative aims to clarify what robots can do and under what conditions they were tested, moving beyond flashy demonstrations that may not accurately reflect performance. Under this new framework, companies will need to publicly disclose the specific tasks their robots can perform, the environments in which they were tested, and any additional training or human assistance required for success. Sunday has identified laundry folding as Memo's first "Solve" capability. The training process for ACT-2 occurs in two stages. Initially, the robot learns fundamental skills from human demonstrations using specially designed sensor-equipped gloves. These gloves, which cost approximately $200 to produce, are intended to closely mimic the shape and functionality of Memo's hands, allowing for precise translation of human movements into robotic actions. After acquiring basic skills, the robot learns from its mistakes using an in-house fleet of robots. Zhao noted that Memo's performance in their office closely mirrors its abilities in real-world settings, thanks to extensive testing in employees' homes and various Airbnb locations. Founded in 2024 by Zhao and co-founder Cheng Chi, Sunday Robotics has quickly grown to a team of over 100 employees. Their focus on the home environment, which differs significantly from controlled factory settings, positions them at the forefront of developing robots capable of achieving physical artificial general intelligence. As various companies prepare to integrate robots into household routines, Sunday is set to launch Memo in a beta program this fall. Zhao emphasized that Memo will operate autonomously, with remote assistance available only when necessary, aiming to build trust with users. While other companies like Weave Robotics and Tesla are also working on home robots, Sunday aims to refine Memo's capabilities, starting with laundry and expanding to tasks like vacuuming and making coffee in the future. The journey towards reliable home robotics is just beginning, and Sunday Robotics is leading the charge.