机器人长出仿生瞳孔：液态金属仿遍猫羊乌贼，强光识别率飙升15%

Core Insights - The article discusses a breakthrough in robotic vision systems, specifically a bionic adaptive pupil reflex visual system developed by teams from the University of North Carolina and Westlake University, which utilizes liquid metal technology to create a dynamic, self-adjusting pupil [2][3]. Group 1: Pupil Functionality - The new system can mimic various animal pupil shapes, such as vertical for cats and horizontal for sheep, allowing for better light regulation and image clarity. It improves image recognition accuracy from 68.38% to 83.56% in bright conditions [3][8]. - Traditional machine vision systems have fixed apertures, leading to overexposure or reliance on computational fixes, which are less efficient compared to the new hardware-based approach [4][12]. Group 2: Liquid Metal Technology - Liquid metal serves dual purposes: it acts as a programmable bionic pupil and simulates neural signals. The system can change shape based on electrical input, allowing for rapid adjustments to light conditions [5][12]. - The design involves eight independently controlled liquid metal units that can transform into various pupil shapes within two seconds, showcasing a significant advancement in hardware capabilities [5][12]. Group 3: Visual Memory and Recognition - The system incorporates a visual memory feature, where repeated exposure to visual stimuli allows for prolonged retention of light signals, akin to human short-term memory transitioning to long-term memory [7][12]. - In tests using the MNIST dataset, the adaptive pupil system significantly enhanced recognition rates under simulated overexposure conditions, demonstrating the effectiveness of hardware intervention over algorithmic solutions [8][12]. Group 4: Practical Applications - The implementation of this adaptive pupil system in robots can enhance their functionality in various scenarios, such as autonomous driving, security patrols, and search and rescue operations, by allowing real-time adjustments to lighting conditions without manual recalibration [12][12].