Restroom Cleaning Robots Secure 500 Million Yuan in Orders

If a robot first learns to clean toilets instead of flipping or serving tea, it may find paying customers faster.

At the 5th Global Digital Trade Expo, a spatial intelligence cleaning robot demonstrated mopping, wiping sinks, and cleaning toilets in a simulated public restroom. The actions are not sci-fi, but the disclosed data is striking: Star Species Robot has 500 million yuan in orders, over 100 devices operating in Hangzhou public restrooms, and received 500+ intent orders at the expo, mostly from overseas.

Meanwhile, garment manufacturing robots repeat a simple task: separating a single piece of fabric from a stack, picking it up, laying it flat, and aligning it. Aitu Technology has signed 200 robot purchase orders this year.

Public restroom cleaning suits robots because the pain points are specific: floors need regular cleaning, sinks accumulate water, toilet stalls are complex, and work is repetitive and labor-intensive. Public venues like airports, malls, transit hubs, and parks may require long hours of cleaning; hiring and scheduling are difficult.

For buyers, it's not a philosophical question but a calculable table: how long a device can operate daily, area covered, staff support needed, recovery time, consumables and maintenance costs, and whether cleaning quality meets standards. This is the commercial value of small scenarios: the task boundary is clear. Robots don't need to understand the whole building, only floors, counters, toilets, obstacles, and cleaning routes. Clearer boundaries mean faster deployment; for clients, easier acceptance and procurement.

Completing a cleaning action at a trade show is not commercialization. Demo environments are tidy, but real restrooms are not: water stains, paper scraps, hair, people passing by, inconsistent dimensions. Robots must not only see but judge stain types, choose actions, avoid pedestrians, and react to low battery or faults.

Garment handling is similar: fabric is soft, easily folded, different materials have different friction, and layers may stick. Aitu's robots need to separate single layers, lay them flat, align to millimeter precision, and handle multiple pieces continuously. The simpler the action looks, the more stability is tested.

So judging commercialization requires five harder indicators: continuous runtime, single-task success rate, human intervention required, fault and maintenance costs, and time to deploy in a similar new site.

Order increases are positive but not all orders are revenue. Three levels: intent orders (interest expressed, but quantity, price, delivery may change; the 500+ at the expo are intent); purchase orders or contracts (more certain but still depend on delivery, acceptance, payment); revenue recognition (after production, delivery, and contract conditions met). Over 100 robots already running means the product has passed the demo stage; 500 million yuan in orders shows demand gathering. But the business model isn't fully proven. Next: can the company produce and deliver on time, is operation stable, can after-sales network keep up, and can orders bring reasonable margins and cash flow.

Focusing on small scenarios doesn't mean embodied AI only does simple work. It's a realistic path to general capabilities. Robots in restrooms accumulate data on stains, spaces, crowds, anomalies; in garment factories, they handle different fabrics, sizes, rhythms. Real data can train models and improve hardware, then expand to adjacent tasks.

After floor cleaning is stable, add counters, walls, and facility inspection; after fabric picking is stable, connect feeding, sewing, and sorting. As more robots enter malls, factories, and public spaces, to judge commercialization, look less at performance videos and more at five tables: operation, intervention, cost, delivery, repeat purchase. These tables are closer to a real business than whether a robot can dance.