旧手机扔了可惜,UCLA团队把它装进双足机器人当大脑

林越2026-10-151990人形机器人

淘汰的旧手机或许不会再沦为抽屉里积灰的电子板砖,而在机器人时代中焕发新的生机。加州大学洛杉矶分校的机器人与机械实验室开发出一款小型步行机器人,其能使用智能手机作为这款双足机器人的躯干、眼睛和大脑。该款机器人的核心理念在于省去了昂贵的额外组件,利用手机的摄像头、惯性测量单元和处理器来处理机器人的视觉、倾斜和运动感知以及关节运动所需的神经网络。这款机器人高48.3厘米,重1.8公斤,包含手机和电池,配备了13个低成本伺服电机,每条腿6个,另有1个用于旋转躯干,使得摄像头可以跟随人移动,而无需整个机器人旋转。

Discarded old phones may no longer become dust-collecting "electronic bricks" in drawers but gain new life in the robot era. UCLA's Robotics and Mechanisms Laboratory developed a small walking robot that uses a smartphone as the bipedal robot's torso, eyes, and brain. The robot's core idea is eliminating expensive additional components, using the phone's camera, inertial measurement unit, and processor to handle the robot's vision, tilt and motion perception, and neural networks needed for joint movement. The robot stands 48.3 cm tall, weighs 1.8 kg including phone and battery, and is equipped with 13 low-cost servo motors—6 per leg, plus 1 for rotating the torso, allowing the camera to follow people without the whole robot rotating.

 

手机并非直接连接到机身,而是通过无线网络与内置的小型电脑通信,由电脑向电机发出指令。研究人员表示,这使得更换不同型号的手机更加便捷。这款机器人目前正在进行动作学习,加州大学团队演示了行走、跟随摄像头拍摄的人以及从俯卧姿势站起等动作。但其还并不能独立进行对话,其语音识别和语言模型都运行在机器人附近的电脑上,通过无线网与手机相连。而步行和视觉识别已经能实现在手机上运行。研究人员在四部安卓手机上测试了PhoneBot,包括一部2017年的荣耀9和三部售价在50到80美元之间的预付费手机。所有手机均测试了核心功能,包括运动感应、行走控制、设备端视觉和交互。

 

The phone is not directly connected to the body but communicates with a built-in small computer via wireless network, which sends commands to the motors. Researchers say this makes swapping different phone models more convenient. The robot is currently learning movements; the UCLA team demonstrated walking, following people captured by the camera, and standing up from a prone position. But it cannot yet converse independently—its speech recognition and language models run on a computer near the robot, connected wirelessly to the phone. Walking and visual recognition already run on the phone. Researchers tested PhoneBot on four Android phones, including a 2017 Honor 9 and three prepaid phones priced between $50 and $80. All phones tested core functions including motion sensing, walking control, on-device vision, and interaction.

 

一个不足之处在于,基于摄像头的地图构建与定位功能在荣耀9上无法运行,研究人员认为这是因为美国的谷歌服务无法在该设备上使用。不过,该论文并未证明PhoneBot能够兼容所有安卓手机。与此同时,这款小型机器人并未配置双臂,其行走训练也仅限于平地。小型电机则限制了这款机器人的负重能力和动作类型。不过,这种相对简单的机器人拥有另一方面的领先优势。研究团队估计,PhoneBot的成本约为400美元,但这不包括智能手机的费用。而论文中列出的其他小型研究用人形机器人价格从3500美元到13800美元不等。

 

One shortcoming: camera-based mapping and positioning failed to run on the Honor 9, which researchers believe is because US Google services cannot be used on that device. However, the paper does not prove PhoneBot is compatible with all Android phones. Meanwhile, this small robot has no arms, and its walking training is limited to flat ground. Small motors limit the robot's payload capacity and action types. But this relatively simple robot has another leading advantage. The research team estimates PhoneBot's cost at about $400, excluding the smartphone. Other small research humanoid robots listed in the paper range from $3,500 to $13,800.

 

PhoneBot的意义不在它能立刻变成消费级产品,而在于它把双足机器人的硬件入门门槛打到了400美元这个量级。前面讨论过万元级人形机器人在硬件上的减配问题,也讨论过双足机器人平衡控制的难度。PhoneBot用一部旧手机替代了昂贵的传感器套件和计算单元,把成本压到了同类型研究机器人价格的十分之一甚至更低。这个思路和前面讨论的用旧手机做大脑、把机器人做成准系统,是同一个方向。它还证明了一件事,机器人的大脑可以当耗材,手机两三年一换,换下来的机型正好可以塞进机器人躯体里。这个模式如果跑通,对机器人开发者生态的意义,可能比产品本身更大。

 

PhoneBot's significance is not that it can immediately become a consumer product but that it pushed bipedal robot hardware entry threshold down to the $400 level. Earlier we discussed hardware downsizing in ten-thousand-yuan humanoid robots and the difficulty of bipedal robot balance control. PhoneBot replaced expensive sensor suites and computing units with an old phone, compressing cost to one-tenth or even less of同类 research robot prices. This approach aligns with the earlier discussion of using old phones as brains and making robots into "barebones systems." It also proves one thing—the robot's brain can be a consumable; phones are replaced every two or three years, and retired models can be inserted into robot bodies. If this model works, its significance to the robot developer ecosystem may be greater than the product itself.