Aisi Technology demonstrates autonomous control of robot dog at Physical AI conference

leaderobot.com reported on Aisi Technology’s exhibition at the 2026 Physical AI and World Model Conference, which opened in Nanjing on Oct. 20, 2026. The exhibition showed a robot dog approaching another robot, greeting it and dancing in response to voice commands.

Aisi Technology’s explanation points to a design that handles models and physical devices within the same operating layer, rather than controlling robots individually. If this setup moves closer to practical use, questions will include not only control of individual machines but also how easily tasks can be designed across multiple devices and updated. Based on what the exhibition showed, the key test is not the visual appeal of the demo but the system’s stability when the same command is repeated and its transferability to different machines. Whether it can meet those tests will distinguish a research exhibit from real-world deployment.

The main point appears to be less the robot dog’s movements than the design linking voice commands, environmental perception, tool calls, execution and feedback on the same screen. According to the published description, Aisi Technology built on its multimodal model with an agent runtime framework that mediates between models and real-world devices, then extended the system to physical hardware through MHS. This can be understood as putting forward an implementation in which a model handles multiple devices while updating task status, rather than controlling a single robot. This newspaper has no related articles, so there is no basis for tracing continuity with past coverage. But based on public information, announcements about robot autonomy are increasing, while many examples remain limited to remote operation or individual demonstrations. In how it was presented, this exhibition also focused on how the components are connected, rather than simply on what the system can do. This suggests that differentiation may depend not only on competition over robot performance, but also on control of the intermediate layers: models, tools, state management and device connectivity. However, the original source is a report conveying impressions of the exhibition; it does not make clear under what conditions the same actions can be reproduced reliably or how broadly the system can be generalized. See the original source for details. Points to watch include, first, which groups of devices MHS supports; second, the safety controls during autonomous execution and the recovery procedures in case of failure; and third, whether the demonstration can be reproduced under conditions close to real-world operations. Confirming these points would make it easier to determine whether the implications of this exhibition are at the research or product stage.