
当投资者正忙于向人形机器人领域投入数十亿美元时,一位拥有三十年机器人研发经验的麻省理工学院机器人专家却指出,此类投资不过是徒劳之举。
扫地机器人Roomba制造商iRobot联合创始人罗德尼·布鲁克斯指出,埃隆·马斯克(Elon Musk)设想的“人形机器人化身万能助手”的未来图景“纯属空想”,部分原因在于,机器人在动作协调性上存在难以突破的瓶颈。
布鲁克斯在博文中写道:“尽管风险投资机构和大型科技企业投入数亿甚至数十亿美元用于训练,当今的人形机器人仍无法掌握灵巧操作技能。”
触觉是人体最复杂的感知系统之一。人类的手部拥有1.7万个低阈值机械感受器,用于感知轻微触碰,且在指尖末端的分布更为密集。这些感受器能对多种刺激(如压力)做出反应,并与15类神经元同步振动。人类试图在机器人中复刻的正是这套复杂机制。
布鲁克斯指出,尽管人工智能已通过大量语音识别和图像处理数据完成训练,但“触觉数据领域尚无此类积累”。他还对马斯克旗下特斯拉(Tesla)及人工智能机器人公司Figure训练人形机器人的方式提出质疑:这两家公司通过让机器人观看人类执行任务视频的方式进行训练,且认为这种方式能大幅提升机器人的操作灵巧度。
布鲁克斯表示:“倘若科技巨头和风险投资机构将投入人形机器人大规模训练的资金缩减至原来的20%,并将全部资金投入大学研究,我认为他们将更快实现目标。”
马斯克曾宣称,特斯拉将在2026年开始销售人形机器人Optimus,且该公司称Optimus已能在特斯拉工厂自主执行任务。与此同时,Figure在今年9月新一轮融资后,投后估值已达390亿美元。但在布鲁克斯看来,所有这些投资最终不过是为人形机器人构建一套极其昂贵的训练体系——而这些机器人终究无法完全复刻人类形态。
布鲁克斯预测,15年后真正能落地的机器人,外形将与人类毫无相似之处——它们可能配备轮子、多条机械臂,或许会保留五指结构的手部设计,但即便如此,这类机器人仍可能被称作“人形机器人”。但就目前的人形机器人研发努力而言,大多数项目将仅存于历史记载之中。
“大量资金将付诸东流,耗费在试图从现有人形机器人身上挖掘出哪怕一丝一毫的性能提升上,但这些机器人终将被淘汰,且大多会被人们轻易地抛诸脑后。”他说道。(*)
译者:中慧言-王芳
当投资者正忙于向人形机器人领域投入数十亿美元时,一位拥有三十年机器人研发经验的麻省理工学院机器人专家却指出,此类投资不过是徒劳之举。
扫地机器人Roomba制造商iRobot联合创始人罗德尼·布鲁克斯指出,埃隆·马斯克(Elon Musk)设想的“人形机器人化身万能助手”的未来图景“纯属空想”,部分原因在于,机器人在动作协调性上存在难以突破的瓶颈。
布鲁克斯在博文中写道:“尽管风险投资机构和大型科技企业投入数亿甚至数十亿美元用于训练,当今的人形机器人仍无法掌握灵巧操作技能。”
触觉是人体最复杂的感知系统之一。人类的手部拥有1.7万个低阈值机械感受器,用于感知轻微触碰,且在指尖末端的分布更为密集。这些感受器能对多种刺激(如压力)做出反应,并与15类神经元同步振动。人类试图在机器人中复刻的正是这套复杂机制。
布鲁克斯指出,尽管人工智能已通过大量语音识别和图像处理数据完成训练,但“触觉数据领域尚无此类积累”。他还对马斯克旗下特斯拉(Tesla)及人工智能机器人公司Figure训练人形机器人的方式提出质疑:这两家公司通过让机器人观看人类执行任务视频的方式进行训练,且认为这种方式能大幅提升机器人的操作灵巧度。
布鲁克斯表示:“倘若科技巨头和风险投资机构将投入人形机器人大规模训练的资金缩减至原来的20%,并将全部资金投入大学研究,我认为他们将更快实现目标。”
马斯克曾宣称,特斯拉将在2026年开始销售人形机器人Optimus,且该公司称Optimus已能在特斯拉工厂自主执行任务。与此同时,Figure在今年9月新一轮融资后,投后估值已达390亿美元。但在布鲁克斯看来,所有这些投资最终不过是为人形机器人构建一套极其昂贵的训练体系——而这些机器人终究无法完全复刻人类形态。
布鲁克斯预测,15年后真正能落地的机器人,外形将与人类毫无相似之处——它们可能配备轮子、多条机械臂,或许会保留五指结构的手部设计,但即便如此,这类机器人仍可能被称作“人形机器人”。但就目前的人形机器人研发努力而言,大多数项目将仅存于历史记载之中。
“大量资金将付诸东流,耗费在试图从现有人形机器人身上挖掘出哪怕一丝一毫的性能提升上,但这些机器人终将被淘汰,且大多会被人们轻易地抛诸脑后。”他说道。(*)
译者:中慧言-王芳
While investors are busy pouring billions of dollars into humanoid robots, an MIT roboticist who has been making robots for three decades claims they are wasting their money.
Rodney Brooks, the cofounder of Roomba vacuum creator iRobot, said the idea of humanoid robots as catchall assistants, the future Elon Musk envisions, is “pure fantasy thinking,” in part because robots are coordination-challenged.
“Today’s humanoid robots will not learn how to be dexterous despite the hundreds of millions, or perhaps many billions of dollars, being donated by VCs and major tech companies to pay for their training,” said Brooks in a blog post.
The sensation of touch is one of the most complex systems in the human body. The human hand contains 17,000 low-threshold mechanoreceptors for picking up light touches, which become denser toward the end of the fingertips. The receptors in your hands respond to myriad stimuli-like pressures—vibrations in sync with 15 different families of neurons. All of this adds up to a complex mechanism that humans want to replicate in robots.
While AI has been trained on large amounts of speech-recognition and image-processing data, “we do not have such a tradition for touch data,” Brooks said, adding he takes issue with the way both Musk’s Tesla and AI-robotics company Figure are training their humanoid robots, with videos of humans performing tasks, assuming this will result in vastly improved dexterity.
“If the big tech companies and the VCs throwing their money at large-scale humanoid training spent only 20% as much but gave it all to university researchers I tend to think they would get closer to their goals more quickly,” Brooks said.
Musk has said Tesla will start selling its Optimus robots in 2026, and the company claims Optimus is already performing tasks in Tesla factories autonomously. Meanwhile, Figure achieved a $39 billion post-money valuation earlier this month after a new fundraise. But in Brooks’ eyes, all this investment adds up to a highly expensive training regime for humanoid robots who will not look exactly like us.
Brooks claims successful robots in 15 years will look nothing like humans—and will sport wheels, multiple arms, and possibly five-fingered hands, though they will still be called “humanoid robots.” But as for today’s efforts, they will largely be relegated to the history books.
“A lot of money will have disappeared, spent on trying to squeeze performance, any performance, from today’s humanoid robots. But those robots will be long gone and mostly conveniently forgotten,” he said.
