也许我们都低估了机器人的发展速度。 昨日，乌克兰第3独立突击旅发布消息称，在哈尔科夫地区，他们首次成功地仅凭无人机和自杀式地面作战平台，迫使俄军士兵向无人装备投降。该通报 称："历史上第一次，俄军士兵在纯粹由机器人系统发起的进攻中投降。"换句话说，机器人的出现，让俄乌冲突发生了前所未有的变化。 本文来自微信公众号：电子工程世界 （ID：EEworldbbs），作者：冀凯、付斌 时间再向前回拨几天，7月2日，消息称乌军的人工智能（AI）战斗机器人已经装备了美国口径为0.5英寸（约1.27厘米）的M2勃朗宁大口径重机枪，目前 正在乌克兰战场上发挥作用。 军用机器人正全面渗透各作战领域，涵盖空中无人机（UAV）、水面无人舰艇（USV）和水下无人潜航器（UUV）。其中，地面机器人（UGV）因面临 的地形最复杂、与环境交互最频繁，成为技术难度最大的领域；无人机发展速度也正在快速发展中。 在地面方面，军用地面机器人类型多、需求大。国外已有不少列装并投入实战，如美国的Pack Bot、TALON等系列。国内暂无明确分类，通常按平台重量 分为微型（≤10kg）、小型（10kg~50kg）、轻型（50kg~100kg）、中型（ ...

Core Viewpoint - The founder of Huawei, Ren Zhengfei, suggests that the Chinese chip industry can break through its current challenges through "stacking and clustering" technology paths, achieving performance comparable to the world's leading levels in computing [2] Group 1: Technological Innovations - China lags behind the US by one generation in single-chip processes, but innovative strategies like "mathematics compensating for physics" and "non-Moore compensating for Moore" can lead to breakthroughs in system-level performance [2] - Huawei's Ascend chip, while not on par with the leading 3nm chips in process technology, utilizes self-developed CCE communication protocols to create efficient clusters, supporting the training of the Pangu large model, achieving computational power comparable to top GPUs [2] - The success of Google's TPU cluster, which trained a 540 billion parameter model, demonstrates the scale effect of cluster computing in AI [2] Group 2: Algorithm Optimization - Huawei's breakthroughs in algorithm optimization are crucial, with the "mathematics compensating for physics" concept reflected in techniques like sparse computing, model quantization, and pruning [4] - The MindSpore framework reduces AI training computational demands by over 30% through dynamic optimization and low-precision computing, showcasing effective software and hardware collaboration [4] - The Tianjin Port's unmanned terminal exemplifies this, where hundreds of Ascend chips form a computing cluster to process massive sensor data in real-time, significantly enhancing operational efficiency [4] Group 3: Chiplet Technology - The "non-Moore" path emphasizes Chiplet technology, which breaks down complex large chips into smaller, functionally distinct chiplets, achieving high-density integration through 2.5D/3D packaging [4] - Huawei's Ascend chip employs a Chiplet architecture, where core computing units pursue advanced processes while I/O, analog, and storage modules use mature processes, achieving performance that rivals or exceeds that of single advanced process large chips [4] - This modular design strategy is similar to AMD's Zen architecture, which has gained significant market share in servers through efficient interconnect technology [4] Group 4: Research and Collaboration - The fundamental bottleneck of China's chip industry lies in basic theoretical research, with Huawei investing 180 billion yuan annually in R&D, including 60 billion yuan for foundational research without performance metrics [6] - Huawei's contributions to game theory, such as the "prophet inequality," address optimization problems, with related papers published at the 65th Computer Science Foundations Symposium [6] - Open collaboration and talent strategies are essential for breakthroughs, with Huawei integrating resources through partnerships with open-source communities and attracting top global talent through initiatives like the "Genius Youth" program [6] Group 5: Industry Outlook - Ren Zhengfei's statements instill confidence in the Chinese chip industry, which has seen numerous companies achieve breakthroughs in mid-to-low-end sectors, particularly in compound semiconductors [6] - Huawei's "stacking and clustering" strategy not only provides a technical pathway for itself but also offers a model for the industry [6] - With ongoing investments in foundational research and the improvement of the industrial ecosystem, the Chinese chip industry is expected to achieve greater autonomy and control in more fields [6]