Tech • AI • Robotics • Game
China Just Unleashed AI Robot Police on Its Streets
7/10TL;DR
China saw two notable humanoid robot trials: Deep Robotics deployed traffic-assistance robots in rainy holiday crowds in Hangzhou, while Archon Robotics showcased kitchen and household automation across multiple humanoid platforms in Shanghai.
KEY POINTS
Humanoids entered daily public work
During the National Day Golden Week, one of China’s busiest travel periods, Deep Robotics placed its DR02 humanoids in Hangzhou’s West Lake area, including scenic entrances, wetland park sites and busy intersections. The test put robots into dense crowds, mixed car and e-bike traffic, and near-constant rain rather than controlled indoor settings.
Traffic duty in real intersections
At roads in the Lingyin area and near the city’s administrative center, the robots directed traffic, gestured to pedestrians, and issued spoken reminders to road users. They were used to spot e-bike riders without helmets, overloaded bikes, pedestrians crossing beyond the line, and cars stopping over the stop line, but they did not issue fines.
Built for bad weather
Deep Robotics said the rain itself was a central part of the trial. The DR02 carries an IP66 protection rating, meaning it is designed to resist dust ingress and strong water jets, and the company used the holiday week to test whether the robot could remain operational through repeated shifts in steady rain.
Adult-sized robot with autonomous movement
The DR02 stands 175 cm tall and weighs 65 kg. It is rated to operate from -20°C to 55°C, walks at up to 1.5 meters per second, climbs steps of about 25 cm, and handles slopes up to 20 degrees. For perception it uses wide-angle cameras, depth cameras and LiDAR, with onboard compute reported at 275 trillion operations per second.
A support role for police and city staff
The robots were designed less as enforcers than as assistants handling repetitive work. By taking over routine warnings and crowd guidance, the system allowed human staff to focus on accidents, serious violations and situations requiring judgment. The company’s stated goal is to move from symbolic robot participation to practical co-working with city personnel.
Tourist information in multiple languages
In the crowded West Lake zone, the robots also answered questions about directions, traffic controls and scenic access in several languages. That effectively turned them into mobile information desks, intended to reduce the time human workers spend repeating basic guidance during peak visitor flows.
A broader push into smart-city services
The holiday deployment remained a trial, with the main aim of evaluating recognition, mobility and interaction in uncontrolled conditions. Deep Robotics framed the longer-term vision as turning humanoids into smart terminals for urban traffic systems and, eventually, wider public-service roles.
Shanghai startup targeted the kitchen
In Shanghai, Archon Robotics, a startup founded in April 2026, presented a debut demonstration of humanoids performing domestic and service tasks. The footage showed robots bending to pick up objects, handling steps, clearing tables, loading or unloading dishwashers, organizing refrigerators and mopping floors.
Team rooted in autonomous driving
The company’s leadership comes largely from self-driving research and driver-assistance systems. Founder and chief robot officer Hongyang Li and co-founder Li Chen are linked to UniAD, which won a best paper award at CVPR 2023, while chief executive Tianyu Li previously worked on Huawei’s ADS 4.0.
Whole-body intelligence across robot types
Archon calls its framework whole-body intelligence. In its description, S1 handles task understanding and multimodal input, while S0 manages balance, coordination and precise movement, functioning as a kind of robotic cerebellum. The demo used hardware from several manufacturers, including modified Unitree G1 units and Booster T2 robots, indicating an effort to make the software work across different humanoid bodies.
Training methods were disclosed, but not metrics
The company said it trains on human demonstrations collected across several Chinese cities, then fine-tunes on real robot data. It also uses corrective examples so robots learn to recover from mistakes such as dropped items or shifting doors. Archon explicitly distinguished between autonomous task runs and teleoperated teaching segments, but it did not release success rates, intervention counts or dataset sizes.
CONCLUSION
The two projects point to a sharper phase in China’s humanoid-robot race: one focused on public streets and city services, the other on kitchens and routine indoor labor. The key question now is not whether humanoids can perform staged actions, but whether they can do ordinary work reliably, safely and at scale.
Ask a question