A shared bike dumped in a flower bed. A construction site nobody is watching after dark. A rooftop a human guard cannot easily reach. In one Shanghai district, the response is no longer just a phone call to a patrol — it is a four-legged machine that walks the block and a drone that flies above it.
In Baoshan District's Nanda Smart City (南大智慧城), six quadruped robot dogs (机器狗) from Unitree (宇树科技) started regular inspection duty alongside drones, forming what local officials call an "air-ground integrated" (空地一体) patrol team. It is one of the clearest signs that China's fast-maturing quadruped robots are moving from trade-show novelty into municipal routines.
What actually rolled out
The deployment began on 28 September 2025. The lineup is specific:
- Three large B1 robot dogs handle bigger outdoor tasks, stepping over curbs and climbing short stairs.
- Three smaller Go2 units slip into narrow green belts and gaps where a person or car cannot go.
- Drones provide the overhead layer, scanning rooftops, construction zones, and other blind spots, and can even drop fire-extinguishing canisters in an emergency.
Behind the scenes sits a partnership struck in February 2025 between Baoshan District, Unitree (宇树科技), and Shanghai University, which set up the Shanghai University General Intelligent Robot Institute (上大通用智能机器人研究院). The institute spent months reworking off-the-shelf robot dogs — adding vision sensors and probes, then training them on local terrain data — before the system could patrol without an engineer walking beside it.
How the patrol works
The dogs are not just remote-controlled toys. According to Shanghai-based CityNewsService and Chinese media coverage, the workflow is a closed loop:
- Detect. Cameras (including infrared and fisheye units) spot problems such as illegally parked shared bikes or piled waste.
- Report. The machine sends the image and location to a central control platform in real time.
- Act. Human security or law-enforcement staff are alerted to intervene, while the robot keeps moving.
- Learn. All patrol data feeds a "Cloud Nanda" governance platform that uses algorithms to generate reports and disposal suggestions.
The value proposition is coverage and frequency: a dog can walk the same route repeatedly, enter spaces people avoid, and keep "eyes" on sites around the clock. Local officials have suggested the mature system could save millions of yuan (≈ US$ hundreds of thousands) a year in security spending, and could be copied in other Chinese cities.
A second front: forests and fire
Shanghai's streets are not the only proving ground. In Beijing's Shunyi District, a roughly 12.5 kg patrol robot dog was brought in for forest-fire duty, fitted with multispectral cameras and a spotlight, running two to four hours per charge, and linked to the district emergency command center. Officials there said the dog can raise patrol efficiency about threefold versus traditional foot patrols, shrinking blind spots in rugged hillside woodland.
The pattern is consistent: cities want machines that extend human senses into places that are tedious, risky, or simply understaffed — and quadruped robots happen to have crossed the threshold of being good enough to try.
Why the partnership shape matters as much as the robot
The Nanda model is the quieter story. A district government, a robot maker, and a university institute each supplied what the others lacked: the city gave a real, bounded deployment environment and the mandate to try; the company supplied the hardware; the institute did the unglamorous work of adding sensors, tuning models, and training on local terrain until the dogs could patrol without an engineer in tow. That three-party shape is reproducible in a way a flashy single gadget is not, and it is why observers should watch the partnership template — district plus maker plus research institute — as the unit most likely to be cloned across Chinese cities.
What to watch
The metrics that will actually tell the story are operational, not promotional: how many routes run unattended for a full year, how often the detect-and-report loop produces an action a human would not have caught, and whether the cost-per-patrol genuinely falls. Privacy norms around persistent aerial-and-ground cameras are the soft constraint that could slow replication more than any technical limit.
Honest limitations
- Reactive, not autonomous policing. Today the dog flags and reports; a human still makes the decision and acts. It is an extra sensor, not a replacement for officers.
- Outdoor ruggedness. As the institute itself noted, many robot dogs were not originally designed for complex outdoor conditions, and the Shanghai units needed significant re-engineering.
- Edge cases. Crowds of curious onlookers, bad weather, and poor GPS under bridges remain hard. Voice warnings and crowd-navigation modules are still described as "future upgrades."
- Privacy and acceptance. Persistent camera-equipped patrols raise questions about surveillance and public comfort that no spec sheet answers.
- Cost at scale. The "millions of yuan saved" claim is a projection, not a measured result, and depends on reliable 24/7 operation that pilots have not yet proven over a full year.
What readers can do now
- If you run a facility or campus: treat robot-dog patrol as a pilot, not a purchase — start with a clearly bounded route (a construction site, a depot, a park) where a closed-loop detect-and-report workflow is easy to measure.
- If you build or integrate: the bottleneck is not the legs but the perception-and-alert stack; the teams winning pilots are the ones who tuned sensors and training data to one real environment.
- If you follow the sector: watch whether the Nanda model (district + robot maker + university institute) gets cloned in other cities — that partnership shape, not any single gadget, is the story worth tracking.