ETH Zurich 将商用机械手改造成靠手指爬行的机器人
Researchers at ETH Zurich have turned a commercial robotic hand into a crawling machine that uses its fingers for movement, balance, and interaction. Powered by an onboard battery and computer, the hand learned its movements through reinforcement learning in simulation. It traversed 14 surfaces, recovered in 21 of 25 fall trials, and correctly executed 29 of 32 keyboard commands after manual alignment. Humans supplied walking and keyboard commands. The researchers envision future robots deploying mobile hands into confined spaces to manipulate objects and press buttons beyond an arm’s reach. But for now, those applications remain a possibility rather than demonstrated capability. Study: “Fingers as Legs: Learning Self-Supported Locomotion and Manipulation with an Anthropomorphic Hand” DOI: 10.48550/arXiv.2609.17172
ETH Zurich 研究者把一只商用机械手加上电池和板载计算机，训练成不依赖手臂、靠手指爬行和保持平衡的机器人。团队在仿真中用强化学习同时训练数千只手，真机在草地、碎石等 14 种表面爬行，25 次被推倒的测试中恢复 21 次，并在弹跳状态下按对 32 次键盘指令中的 29 次。
原文给出爬行、跌倒恢复和按键三组具体数字，适合用视频画面配合数据做一次效果展示。
粉丝 8万
The latest in AI and robotics, explained Founder & CEO of The Rundown AI Zurich 🇨🇭 / NYC 🇺🇸
- 浏览
- 1.1万
- 点赞
- 587
- 评论
- 12
- 分享
- 22
- 收藏
- 59
数字来自平台，为最近一次抓取时的数据。
以下是视频口播的自动转写，可能有错字。
So, this robotic hand walks on its fingers. Researchers at eth zurich took a commercial robotic hand, added a battery and an onboard computer, and trained it to move without an arm.
But the catch is, every time a finger takes a step, the others have to keep the entire robot standing. And those fingers have different lengths, with a thumb pointing off to one side.
So, the team used reinforcement learning, training thousands of simulated hands at once. Each fingertip has its own reference position, but the system learned when to lift each finger and to take a step.
The real hand crawled across 14 surfaces, including grass and gravel. And when knocked over, it recovered in 21 out of 25 trials.
It also pressed keyboard keys while bouncing on its remaining fingers, hitting the requested arrow key 29 out of 32 times.
And yes, it does look disturbingly like thing from the addams family, but the researchers envisioned some practical use cases for this in the future.
Like maintenance robots, for example, could one day send a hand through a small access hatch to press a button or move a loose part inside machinery where its whole arm can't fit.
If you want to stay in the loop with the latest in ai and robotics, follow for more.
来源：TikTok · Rowan Cheung · tiktok.com