Core Viewpoint - Beijing Mianbi Intelligent Technology Co., Ltd. has recently published a patent for a "Dynamic Sparse Activation-Based Human-like Robot Multimodal Perception Fusion Method" which aims to enhance robot functionality through advanced perception techniques [1] Group 1: Patent Details - The patented method involves acquiring multimodal data streams from the robot and generating modality availability mapping based on the robot's real-time operational status [1] - It employs a sparse gating strategy to dynamically select which perception modalities to activate, constructing an active modality set based on the selection results [1] - The method utilizes a Kalman filtering algorithm to evaluate the state of the active modality set, identifying modality switching states [1] Group 2: Operational Enhancements - Upon recognizing modality switching, the method constructs a state evaluation window based on task complexity and progress changes, dynamically adjusting the robot's state space dimensions [1] - This approach generates a multimodal state set that informs the robot's task execution instructions and controls its behavioral actions [1] - The innovation addresses the issue of mismatched state space dimensions during modality switching when using the sparse gating strategy [1]