Post by Wish Lab

A new robotics company, Sunday, founded by Tony Z. Zhao and Cheng Chi has unveiled a breakthrough that may mark a paradigm shift in robotics. Their new robot, ACT-1, is trained purely on human data — not teleoperation data, not synthetic data, not robot-collected trajectories. This is a huge leap forward for robotics, a field that has always suffered from a lack of large-scale, high-quality motion data. Why it matters: Robotics has been stuck in a "data deadlock." You need millions of real-world examples to train general-purpose robots, but you can’t collect that data without already having capable robots in homes. ACT-1 breaks this loop. Sunday’s key innovation: 🔹 Skill Capture Glove — a human-wearable glove that mirrors the exact geometry and sensor layout of their robot hand. If a human can do it, the robot can too. 🔹 Skill Transform — a method that converts human motion directly into robot data with 90% accuracy, eliminating the embodiment mismatch. 🔹 Memo, their robot platform — designed from the ground up for dexterity, force sensitivity, and whole-body mobility. ✅The result? ACT-1 shows zero-shot generalization to completely new environments — including unfamiliar Airbnb homes — and performs long-horizon tasks like clearing a dining table, loading a dishwasher, navigating rooms, folding socks, and even pulling an espresso shot. This may genuinely be one of the most important steps toward putting a robot in every home. #Robotics #ArtificialIntelligence #ACT1 #SundayRobotics #RobotAI #FoundationModels #DeepLearning #Innovation #TechNews #MachineLearning #Autonomy #FutureOfWork #HomeRobotics #AIResearch #ComputerVision #HumanData #RoboticsInnovation #AITrends

Video Content