在第26届中国国际工业博览会上,复旦大学研发的跑腿机器人“旦宝”正式发布。该机器人搭载纯视觉端侧大脑,无需预建地图或遥控,能在复杂环境中自主通行,标志着机器人自主化范式的重大突破。
AI-generated summary
传统机器人主要依赖遥控、预编程或预先建图,缺乏对环境的实时自主理解能力。
中新社上海10月12日电 (记者 陈静)12日,在第26届中国国际工业博览会(工博会)上,搭载纯视觉机器人端侧大脑的跑腿机器人“旦宝”正式发布。
复旦大学未来信息创新学院博士后张天资介绍,纯视觉机器人端侧大脑技术突破让机器人摆脱对预建地图等的依赖,将推动机器人自主化底层范式从“依赖预设”走向“自主理解”。“过去的机器人按预设或建图行动,未来的机器人要靠智能理解世界。”
长期以来,机器人的运动与作业主要依靠三种方式:遥控、预编程和预先建图。在遥控之外,无论是预先设定的程序轨迹,还是预建地图,都要在运行之前给定,而非来自机器人自身对环境的实时理解。一旦环境或任务发生变化,就需要重新编程或重新建图。
据了解,仅凭视觉感知,在功耗不到20瓦的端侧芯片上实时完成空间理解与自主决策,“旦宝”即可在复杂、陌生的环境中自主通行,无需遥控,也无需预建地图。端侧大脑让机器人直接从视觉中理解所处的空间,并据此自主决策与行动,不再依赖实时遥控、预设程序或预建地图。
研发团队正与跑腿配送、巡检、商用清洁等多类机器人开展适配,同时在国产端侧算力芯片上并行适配。团队将沿着“空间智能-任务智能-端侧通用智能”的路径,让机器人从自主行走,走向自主完成任务,再到理解意图、持续进化。
AI outlook — possibilities, not facts
研发团队将持续在跑腿配送、巡检等领域进行适配测试。
Very likely · Within months

Wang Xingxing, CEO of a Hangzhou-based robotics company, suggests that recent developments in OpenAI's GPT models could accelerate the timeline for achieving breakthroughs in embodied intelligence to within one or two years.
The 26th China International Industrial Expo opened in Shanghai. The "Qingshi Intelligent Drone Inspection Software System" developed by East China Normal University realizes drone coordinated inspection and precise positioning; Fudan University demonstrated an autonomous desert planting robot that collaborates with air, space and ground to improve desertification control efficiency and large-scale operation capabilities.

Zhejiang Province has recently held the World Artificial Intelligence Open Source Competition, the Intelligent Robot Application Skills Exhibition and the Global Digital Trade Expo, and has made efforts in the three dimensions of basic capabilities, application implementation and global open cooperation, aiming to create a global artificial intelligence innovation and development highland.

Microsoft President Smith called for the establishment of a global AI regulatory framework and independent review agency, emphasizing the insufficiency of current AI safety regulations, suggesting the introduction of emergency braking mechanisms and incident reporting systems, and affirming the UK's leading position in AI safety research institutes.

At the 26th China International Industrial Expo, the "Danbao" robot equipped with a pure visual robot end-side brain was released. This technology enables robots to get rid of their dependence on preset maps and remote control, and realize independent understanding of the environment and decision-making. It is lightweight and has low power consumption, providing a path for the large-scale implementation of embodied intelligence.

China accounts for about 5 per cent of equipment and materials directly supplied to the US chip industry by country of manufacture, but holds over a 90 per cent grip on necessary rare earths, according to the Semiconductor Industry Association.