Robots that understand the physical world Zhicheng AI builds the Chengling World Model, an end-to-end neural network that lets robots predict outcomes before they act — moving from seeing the world to understanding, predicting and changing it. The endgame of general AI isn’t a smarter chat window. It’s a robot that works and lives alongside us. Z Robotics Inc (“Zhicheng AI”) was founded in March 2024 to focus on embodied intelligence. We design and train an AI-native physical-world model end to end, giving robots stronger generalization in real environments. TR and XR share our in-house Chengling physical-intelligence brain. Form follows the task while intelligence transfers: XR serves industry today, TR is built for human environments tomorrow. TR6 General-Purpose Humanoid A bionic bipedal design with hybrid series-parallel actuation, built for spaces made for people: stairs, thresholds, narrow corridors. XR1 Wheeled Dual-Arm Robot Dual-arm manipulation, omnidirectional mobility and a lifting torso. Modular and ready out of the box for assembly, inspection and handling. Rehearse the outcomebefore making a move The hard part isn’t seeing — it’s predicting. Chengling models state, causality and the consequences of actions together, cutting costly trial and error in the real world. In the lab: capability.In the real world:mastery. Light changes, objects get occluded, tasks change mid-stream. XR1 grasps, hands over and collaborates in the unstructured settings of trade shows and production lines. Start where value is proven Robots aren’t here to replace people, but to free them from dangerous, repetitive and draining work. Open research,verifiable progress From spatial visual perception to cross-embodiment manipulation, we publish our core methods. 2025.10EBT-Policy: Energy Unlocks Emergent Physical Reasoning CapabilitiesarXiv:2510.27545 ↗ 2025.09Tenma: Robust Cross-Embodiment Robot Manipulation with Diffusion TransformerarXiv:2509.11865 ↗ 2025.05Spatial RoboGrasp: Generalized Robotic Grasping Control PolicyarXiv:2505.20814 ↗ 2024.11Spatially Visual Perception for End-to-End Robotic LearningarXiv:2411.17458 ↗ 2024.09Generalized Robot Learning FrameworkarXiv:2409.12061 ↗ Working with industry and research partners Bring us your hardestreal-world problem Our Chengling World Model lets robots predict outcomes before they act — from seeing the world to understanding, predicting and changing it. Founded in Hangzhou in March 2024, Zhicheng AI designs and trains an AI-native physical-world model end to end, giving robots stronger generalization in real environments. Rehearse the outcome before making a move Chengling models state, causality and the consequences of actions together, cutting costly trial and error in the real world. 01PerceiveLanguage · RGB-D vision · touch and force · joint trajectories 02UnderstandSpace · object relations · dynamics · task intent 03PredictFuture states · consequences · what can be manipulated · uncertainty 04ActPosition · direction · frequency · speed · force feedback In the lab: capability.In the real world: mastery. Bring us your hardest real-world problem