Sunday Robotics ACT-1

Sunday Robotics' 2025 household robot model, trained with not a single teleoperated robot demonstration in its data.

ACT-1 is the first foundation model announced in November 2025 by Sunday Robotics, a US home-robotics startup founded by ALOHA co-creator Tony Zhao and Diffusion Policy co-creator Cheng Chi. Its training data contains no teleoperated trajectories at all: a person wears a “skill-capture glove” matching the robot hand's shape and sensor layout, does household chores while wearing it, and Skill Transform then strips out the human-specific parts of the footage and motion, converting it into robot data (the company reports roughly a 90% conversion success rate) — sidestepping the fact that teleoperation is slow, expensive, and hard to scale. ACT-1 drives the household robot Memo, using a single end-to-end model for both long-horizon manipulation and navigation via a 3D map; demos include clearing a table into a dishwasher, folding socks, and making coffee, plus tidying a table in an Airbnb it had never visited. In July 2026 the company previewed ACT-2.

ExampleIn the “table to dishwasher” task, ACT-1 has Memo clear wine glasses, ceramic plates, and metal utensils from a table, scrape off leftovers, load the dishwasher, and start it — 33 kinds of dexterous interactions, 68 in total, across 21 different objects, over more than 130 feet (about 40 meters) of travel.

Also called

ACT-1: A Robot Foundation Model Trained on Zero Robot Data, ACT-1 (act one)

Related

Skill Capture Glove · Sunday Robotics Memo · Sunday Robotics · Robot-free (Embodiment-free) Data Collection · Mobile Manipulation · Long-horizon Task

Sources

ACT-1: A Robot Foundation Model Trained on Zero Robot Data (Sunday blog, 2025-11-19)
Sunday Robotics 公司页（创始人介绍） (Chinese)
ACT-2 Preview: Generalizing Reliability (Sunday blog, 2026-07-17)

As of

2026-07