Amap, Alibaba's mapping and navigation platform, has put its quadruped robot dog Tutu to work on routine inspection rounds at Alibaba's Xixi campus in Hangzhou. During Alibaba's 2026 Qincheng Conference, Amap's embodied AI team released a smart campus case study and joint solution with Alibaba Group's general affairs department and the campus services unit Qincheng Zhihui, according to a report by the Chinese robotics outlet Robot Qianzhan published on Tencent News.
Amap is not building its own robot hardware or taking over campus operations. Instead, it supplies spatial intelligence and task execution: it calls on a robot body to move, and connects upward to the campus task and operations systems. Over recent months Tutu has been tested in guiding visually impaired people, tours, shopping assistance, inspection and last-mile delivery.
The software behind it is Amap's ABot system, organized in three layers. The data layer, ABot-World and ABot-Earth, provides a world model and spatial data base for simulation and environment understanding. The model layer includes the navigation foundation model ABot-N1 and the manipulation model ABot-M0.5. The agent layer, ABot-AgentOS and ABot-ER, handles task planning, execution and memory, and tracks object states and spatial relationships. Movement uses a two-speed design: the cloud does high-level planning from task and map data, while the robot generates its path on the device from live camera input, so it can adjust around narrow corridors, crowds and temporary obstacles instead of stopping.
What is new at Xixi is the work-order loop. The joint solution turns a service catalog, inspection targets, frequency and acceptance standards into executable tasks, then connects them to maps, access permissions, campus facilities and work orders. For a fire shutter door, for example, the robot must reach the right spot, adjust its viewing angle so the photo matches the inspection requirement, and send the evidence back for staff to review and act on. Difficult cases are reviewed and labeled by people, then fed back into model evaluation and testing. Ad hoc jobs, such as checking a temporary construction site, can be inserted and the robot then resumes its regular route.
The roles are split four ways: Alibaba's general affairs team sets real tasks and standards, Qincheng Zhihui runs service operations and acceptance, Amap provides navigation and multi-step execution, and campus professionals make final decisions. Security patrol is the first use case, with light delivery, reception tours and fire safety support planned next.
The open question is whether the same stack can be reused across other campuses while keeping each task's standards and lines of responsibility.
SEE ALSO: Amap Open-Sources Lambda-0 · Amap Releases ABot-Earth 0.7