The Beijing Humanoid Robot Innovation Center has recently announced a new patent that enables robots to possess zero-sample self-reflective capabilities. As reported on February 14, 2026, by Renmin Caixin, the company has applied for the patent titled “Robot Zero-Sample Self-Reflective Task Planning Method and System.”
The patent summary indicates that this innovation involves a method and system for zero-sample self-reflective task planning in robots. It includes the following components: using a zero-sample perception module to analyze colored images of a scene, extracting scene feature descriptions, object detection boxes, and pixel-level mask information; feeding the scene feature descriptions and task instructions into a task planning model to generate an initial task plan, which is then refined into an advanced task plan by a reflective judgment model; identifying target object detection boxes based on current sub-tasks; and utilizing a target tracking model to predict object locations and assess their motion status.
When a target object is stationary, the method generates candidate grasping postures based on its colored depth image using a grasp detection model. A visual language model then suggests a grasping area, and the optimal grasping posture is selected through evaluation metrics. Following the execution of the sub-task, the reflective judgment model analyzes the task status; if successful, the process continues, and if not, it conducts analysis and planning corrections until task completion. This method endows robots with zero-sample self-reflective abilities, enhancing the robustness of task execution.
For continuous updates on market dynamics and policy insights, you can download the official Securities Times app or follow their official WeChat account.
Original article by NenPower, If reposted, please credit the source: https://nenpower.com/blog/beijing-humanoid-robot-innovation-center-unveils-patent-for-zero-shot-self-reflection-capability-in-robots/
