Sunday Robotics Unveils ACT-2 Preview for Home Robots

Sunday Robotics unveiled ACT-2 Preview and used its Memo robot to showcase household manipulation in real homes. The release drew attention because the company framed it not as a one-off demo, but as evidence that a home robot can operate more reliably in unseen environments—a central obstacle for real deployment.

Key results

According to reposts summarizing the release, ACT-2 achieved a 99.1% zero-shot success rate on laundry folding across 785 trials in 31 unseen environments. The reported test scope covered nine garment categories, sizes from XXS to 8XL, and variations in lighting, surfaces, and initial placement. Folding quality was described as near human level, with an average score of 4.72/5 and 98.3% of samples receiving 4 stars or higher.

Model claims and evaluation

Sunday Robotics described ACT-2 as a model meant to unify broad generalization and high reliability. Multiple posts said the company claimed Memo could learn a new behavior from just one fine-tuning example and then generalize it to unseen real homes. The company also introduced an evaluation standard called Solve; reposted summaries said its core idea is to define the task scope up front and then test in unseen environments, rather than relying only on controlled demos.

Reactions and limits

Commentary in the cluster treated the demo as a notable step for home robotics, with some posters calling the result close to “out of the box.” At the same time, nearly all details here come from Sunday Robotics’ own release materials and reposted interpretations, so outside discussion mainly centered on the reported metrics, the evaluation framing, and whether this level of household generalization will hold beyond the showcased tasks.

2026-07-17 ~ 2026-07-18 · 17 related posts

Full story(2 episodes)→

5 near-duplicate retellings: tonyzzhao · tonyzzhao · tonyzzhao · tonyzzhao · tonyzzhao