Flow - 4x faster than typing
Mini toolsACT 2
By Sunday
ACT 2 is a robot manipulation model built on a pretrained base trained from sensorized human demonstration data. Its post training process can learn new, generalizable behaviors from a single demonstration and iteratively close reliability gaps found during real world deployment. Running on Sunday Robotics mobile home robot Memo, ACT 2 was evaluated on laundry folding across 785 autonomous attempts spanning 9 garment types in unseen homes, achieving a 99.1% success rate and a mean fold quality of 4.72 out of 5 stars, with zero home or garment specific adaptation. The model shows emergent behaviors including recovery of dropped or crumpled garments, robustness to physical disturbance, and manipulation across a wide range of garment sizes. The same base model is also being extended to other household tasks including vacuuming, toy organization, zipper fastening, and coffee preparation.
New
Multimodal
Overview
ACT 2 is a robotics foundation model developed by Sunday Robotics for autonomous home manipulation. Pretrained on a large, high quality sensorized human demonstration dataset, it generalizes learned behaviors to unseen home environments with minimal additional data. In evaluation it achieved a 99.1% zero shot success rate folding diverse garments across unfamiliar rooms, surfaces, and starting configurations, with no per home data collection or fine tuning required.
Last updated: July 17, 2026
