- Gatsby launched the first US consumer humanoid robot cleaning service on May 14, 2026, charging $150 per apartment clean and operating as a hardware-agnostic platform rather than a robot manufacturer.
- Figure AI’s Helix 2.5 achieved a 56% task success rate across 30 unfamiliar homes in September 2026, with bed-making succeeding 67% of the time but toy-tidying dropping to 40%.
- The International Federation of Robotics recorded roughly 7,000 humanoids in industrial and professional service use in 2025, against tens of millions of robot vacuum units shipped, a gap that illustrates how far consumer humanoid deployment has to travel.
A humanoid robot cleaned a paying customer’s apartment in the US for the first time on May 14, 2026. The company behind it, Gatsby, is not building the robots: at $150 per clean through an iOS app, it is positioning itself as a hardware-agnostic dispatch layer, sitting at the low end of what professional human cleaners charge. The commercial model exists. Whether the hardware can sustain it is a different question entirely.
What the Hardware Actually Does
Cleaning an unfamiliar apartment autonomously demands real-time scene understanding, precise manipulation of objects the robot has never encountered, and enough on-board compute to make decisions faster than the environment changes. Figure AI’s Figure 03, powered by its Helix 02 vision-language-action model, demonstrated living room cleanup in March 2026, handling complex two-handed behaviour like lifting a bin while scooping blocks and stashing items under one arm. The robot integrates cameras that Figure says double frame rate and cut latency by 75% compared to its predecessor, according to the company, plus custom tactile sensors sensitive enough to register the weight of a paper clip. That sensor resolution matters: a robot that cannot feel what it is holding will crush a wine glass or drop a phone.
Tesla’s Optimus Gen 3, launched in 2026, takes a different approach to manipulation.
The 44% Problem
Tasks failed 44% of the time. That is the inverse of Figure AI’s Helix 2.5 result: a 56% complete-task success rate across 30 unfamiliar homes in September 2026, up from a 9% experimental baseline. Performance split sharply by task type. Making beds succeeded 67% of the time and folding towels 62%, but tidying toys dropped to 40%. Repeatable, geometrically predictable tasks are tractable; general-purpose object handling in a cluttered room is where the system breaks down. That gap is exactly where domestic deployment lives.
1X’s NEO, targeting 2026 deliveries, is expected to launch with limited autonomy and an ‘Expert Mode’ for remote supervision on unfamiliar tasks, a more candid framing of where the technology actually sits.
From Warehouses to Living Rooms
Agility Robotics’ Digit has been operating in customer facilities, moving totes and stacking boxes in structured logistics environments. Living rooms are harder. Adapting hardware for unstructured domestic space means more sensitive actuators, force-feedback sensors rated for fragile objects, and compute powerful enough for real-time inference but efficient enough to run on battery for a full shift. The deployment challenges for humanoid robots in 2026 go beyond any single hardware specification. Tesla’s deployment of Optimus units for autonomous floor cleaning at Gigafactory Texas since April 2026 is the clearest real-world durability test so far, though a factory floor is still a long way from an apartment in the Mission District.
Who Can Actually Afford One
The International Federation of Robotics recorded roughly 7,000 humanoids in industrial and professional service use in 2025, against tens of millions of robot vacuum units shipped in the same year. Tesla has indicated Optimus Gen 3 consumer pricing in the $20,000 to $30,000 range, though timelines and rental programme details remain unconfirmed. At those numbers, mass adoption depends on the robot reliably doing enough to justify the spend, a bar the current success rates do not yet clear.
Reliability at Scale
Figure 03’s operational endurance remains an open metric: the company has not published sustained-use figures for multi-hour residential shifts.
Goldman Sachs forecasts a $38 billion humanoid market by 2035. The hardware being built today, tactile sensors, tendon-driven hands, vision-language-action models running on custom silicon, is what that forecast is riding on. Gatsby’s first paying customer is a data point. A 56% task completion rate across unfamiliar homes is another. Neither number is good enough yet.
