Today, we present a step-change in robotic AI @sundayrobotics. Introducing ACT-1: A frontier robot foundation model trained on zero robot data. - Ultra long-horizon tasks - Zero-shot generalization - Advanced dexterity 🧵->

Instead of teleoperation, we train solely on data from our Skill Capture Glove. The glove is co-designed with Memo's hand, meaning they share the exact same geometry and sensor suite. If you can do it wearing the glove, Memo can learn it.

The Skill Capture Glove gives us two orders of magnitude higher capital efficiency compared to teleoperation ($200 vs $20,000) It also allows us to scale diversity faster. You can collect data anywhere without needing to move robots around.

Skill Capture Glove aligns the hands, but what about the rest of the body? Human collectors vary in height and arm length, and are also visually different. We developed Skill Transform, a method that converts glove data into equivalent robot data with a 90%+ success rate.

It took us over a year to engineer the core infrastructure. We then spent the past 3 months to produce all the autonomous results above. Below, I highlight some of my favorite parts of this release.

The table-to-dishwasher task is the classic nightmare scenario for roboticists: Long-horizon, highly dexterous, precise, whole-body manipulation combined with delicate, transparent, reflective, and deformable objects. Yet Memo handles it so naturally and elegantly.

Specifically, wine glass loading is the most delicate subtask: Push down with too much force? Shatter. Insert the wrong prong? Shatter. We broke many during development, but zero over 20+ live demo sessions.

One less-known fact about glove-based data collection: it produces higher quality data than teleop on contact-rich tasks. Remote teleop can’t provide good force feedback, but gloves do naturally, making tasks like sock folding, which rely on feel, far easier to capture.

It is even more fun to see how Memo reacts to unseen environments. We deploy it to 6 unseen Airbnbs and task the robot with fine-grained tasks such as picking up utensils from the plate. Because we train on data from over 500 homes, the new home is instantly familiar to Memo.