Zhiyuan Innovation unveils AGILE2.0 perception-control model, emphasizing whole-body control in dynamic environments

Zhiyuan Innovation (Shanghai) Technology Co., Ltd. unveiled the AGILE2.0 perception-control integrated model (AGIBOT Generative Intelligent Locomotion Engine) on September 11, 2026. The company presented the concept of robots “seeing, thinking and moving” simultaneously, and demonstrated the capabilities in a creative short film, “Acrobatic BOT.” It said the Lingxi X2 performed movements including passing through a ring of fire, spinning a diabolo, three-person long-rope jumping, two-person vaulting, two robots working together to climb onto a raised platform, juggling three balls and collaboratively carrying boxes.

Zhiyuan’s announcement points to an effort to extend robot “see and move” control beyond static demonstrations and into dynamic environments. Passing through a ring of fire, jumping rope and carrying boxes, cited in the announcement, are scenarios that require visual, balance and contact control at the same time, offering a basis for assessing the conditions under which the system could be used in real machines.

AGILE2.0’s novelty lies not simply in improving locomotion or grasping individually, but in handling the process from visual perception to whole-body motion control as an integrated closed loop. Zhiyuan has said it advanced the connection between perception and motion with AGILE1.0. It positions AGILE2.0 as bringing together the end-to-end process, including environmental observation, terrain understanding, dynamic traversability analysis and contact-state switching. In other words, it extends an existing direction, while shifting the implementation focus from “can move” to “can move while observing and responding to changing conditions.” In a related article, [Zhiyuan launches 2027 graduate recruitment at University of Electronic Science and Technology of China, to jointly build an industry-academia collaboration base](/news/satoru-moto-denshi-kagidai-gaku-de-2027-todoke-kou-shou-wo-k-68f72e), this newspaper reported that the company planned to jointly establish an industry-education integration and innovation practice base for embodied intelligence with the university, advancing equipment sharing, joint curriculum development, collaborative research and joint practical training. AGILE2.0 points to a continuum between that talent and research infrastructure-building and the strengthening of control models on real machines. However, graduate recruitment and the joint establishment of a base concern talent supply, while AGILE2.0 concerns control architecture; even within “embodied intelligence,” they operate at different layers. They should not be conflated. From an industry-structure perspective, AGILE2.0 highlights the degree of integration between perception and control in real environments, rather than a competition over robot hardware performance. Ring-of-fire crossings, rope jumping and box carrying are showy movements, but the focus is not the performance itself; it is how well continuous control can be maintained in dynamic environments. If the approach proves stable, the basis of competition could shift from hardware alone to the integration of real-world data, motion learning and control. The announcement, however, does not specify operational uptime, recovery conditions after failures, the scope of tasks covered or how often the model is updated. The next focus will be determining in which environments, and to what degree of autonomy, the system can operate, as well as under what conditions human intervention is needed.

The company said AGILE2.0 is an end-to-end integrated mechanism that deeply combines environmental observation, terrain understanding, dynamic traversability analysis, whole-body motion control, contact-state switching and fine end-effector manipulation. It said the model handles visual perception, environmental understanding, whole-body motion control, and coordinated upper- and lower-limb operations in a single closed loop, without intermediate modules or artificial stage divisions. The company also said that, building on AGILE1.0, the system can handle dynamic obstacles and environments with substantial changes, collaboration among multiple robots, and safety protection and autonomous recovery when people enter the area unexpectedly. The announcement described the model as a foundation for motion intelligence aimed at scaled deployment.

In Japan, humanoid and industrial robots have strengths in mechanical precision and deployment at worksites. In an integrated approach such as this one, combining visual perception and whole-body control, collecting real-world data and building a learning infrastructure could become key issues. AGILE2.0 points to a competitive landscape in which linking data collection, learning and control matters more than standalone performance, raising the importance of design capabilities that encompass learning environments as well as components and robot bodies.