- AI-powered robot based on M20S completed live rescue drill in Zhejiang mine
- System combines robot dog, satellite links and large AI models
DEEP Robotics (云深处科技), together with a Zhejiang government agency and a telecom operator, unveiled on May 11 what they described as the industry’s first rescue communication robot for confined spaces.
As part of the launch, the Hangzhou-based developer of quadrupedal and humanoid robotics completed a live search-and-rescue drill inside a mine tunnel in Quzhou, a city in western Zhejiang, to test the system in a high-risk environment.
The system, developed together with Zhejiang’s emergency management department and China Unicom’s Zhejiang branch, targets rescue operations in high-risk environments such as mines and tunnels, where unstable terrain, flooding and communication blackouts often hamper emergency response efforts.
The robot combines a quadruped platform, AI models and 5G-plus-satellite communications to enter hazardous areas without human rescuers. It can conduct reconnaissance, assess risks, maintain communications and search for trapped people.
DEEP Robotics leveraged its Lynx M20 wheeled quadruped robot as the platform. The machine can climb 45-degree slopes, cross obstacles up to 80 centimeters high and move through water deeper than 40 centimeters.
The robot carries thermal imaging systems, life detectors, gas sensors, mesh-network communication modules and loudspeakers.
During the live drill, it identified falling rocks and flooding hazards, triggering a yellow alert. After detecting signals from trapped individuals, the system escalated the warning to red and activated thermal imaging and life-detection functions.
The platform uses a hybrid AI architecture in which smaller front-end models are tasked with processing on-site imagery in real time, while larger cloud-based models perform deeper analysis and generate response recommendations.
The system can automatically issue red, orange and yellow alerts based on danger levels.
The developers said they trained the system using more than 12,000 real disaster-response data samples covering flooding, leaks, falling debris and trapped personnel scenarios, aiming to shift rescue work from reactive response toward predictive risk assessment.
The launch follows an earlier flood-response robot system jointly developed by DEEP Robotics and Zhejiang emergency authorities.
