By Dan Smigrod, Managing Editor, Humanoid Home News
Source: Sunday Robotics
Context: Manufacturer research release reporting ACT-2 laundry-folding evaluations in unseen homes.
Published: July 17, 2026
Sunday Robotics says its Memo home robot completed 778 of 785 autonomous laundry-folding attempts using ACT-2, producing a reported success rate of 99.1% across unfamiliar household environments.
The company describes the evaluation as zero-shot because it used one fixed model checkpoint and no home-specific or garment-specific data, demonstrations or post-training at deployment. Sunday says none of the evaluation homes or physical garments were used for task-specific post-training or model selection.
The claim has a defined boundary
The 785 attempts covered nine garment groups, including T-shirts, pants, leggings, shorts, polos, blouses and sleeveless or long-sleeved tops. Garment sizes ranged from XXS to 8XL, with varied colors, materials, textures and starting positions.
Sunday excluded socks, bras, underwear and accessories because those items are commonly paired, sorted or hung rather than folded. The evaluation also varied rooms, beds, folding surfaces, lighting, robot positions and whether clothing began in a basket, a pile, on a bed or on the floor.
An attempt counted as successful when Memo folded the garment and autonomously added it to a stack. Sunday reports 99.1% overall success with a standard error of 0.3 percentage points. T-shirts accounted for 312 attempts. Blouses had the lowest category result at 94.7%, while several categories reached 100% in the reported sample.
Quality and speed were measured too
Sunday says its annotators graded completed folds using a five-star rubric fixed before evaluation. A second annotator independently reviewed each initial grade, with disagreements resolved by a review lead.
Across the 778 completed folds, the reported mean quality score was 4.72 out of 5. The company says 98.3% received four or five stars, including 73.8% rated at five stars. Successful folds took a median of 2 minutes 13 seconds and a mean of 2 minutes 19 seconds, including autonomous retries and recovery.
Sunday calls laundry its first “Solve,” a proposed reporting framework that combines performance, the scope of tested conditions and the adaptation required at each deployment. The company says ACT-2 is also learning vacuuming, toy organization, zipper fastening and coffee preparation, but those behaviors have not yet met the same declared standard.
Manufacturer evidence still needs independent confirmation
The detailed sample size, task boundary and published grading method make the claim easier to evaluate than a short demonstration video. Sunday also links to videos of the attempts. However, Sunday designed the study, operated the robots and graded the results. The reported performance has not been independently replicated, and it does not yet establish reliability in ordinary customer use.
Sunday says it plans to deploy Memo with families through a beta program in fall 2026. That step should provide a more consequential test of maintenance, user interaction and performance outside a company-run evaluation.
Humanoid Home News takeaway: Sunday’s 785-attempt result is notable because it states what counted as success, what objects were excluded and what per-home adaptation was required. The next question is whether Memo can sustain comparable reliability during unscripted household beta use.
