一袋零食，难倒11支顶尖机器人团队

Core Insights - The AgiBot World Challenge at IROS 2025 highlighted the challenges faced by teams in completing tasks, particularly the "hanging snacks" task, where no team succeeded in executing the full sequence of actions [1][2][5] - The competition emphasized the importance of model generalization and stability across multiple scenarios rather than just algorithm performance [1][10] Task Performance - The highest completion rate for the "hanging snacks" task was only for the action of grabbing the snack bag, indicating significant difficulties in executing the complete task [2] - In contrast, teams performed well in other tasks such as folding clothes and pouring water, suggesting advancements in pre-training and imitation learning for medium complexity tasks [6][10] Technical Challenges - The "hanging snacks" task required high precision in visual recognition and spatial positioning, complicated by environmental factors such as color similarity and the small size of hooks [5][10] - The competition revealed that while algorithms can achieve "can it be done" status, the transition to practical application involves significant engineering challenges [10][11] Industry Developments - The event served as a systematic validation of embodied intelligence infrastructure, showcasing the need for a complete chain from data collection to model deployment [11] - Following the competition, a new secondary development platform called "LingChuang" was launched, aimed at lowering barriers for developers through AI visual action extraction and cloud-based learning frameworks [11]