Robots that learn from human demos can’t tell how hard to grip.
I made a “force sensor” with no added electronics, using only an iPhone to read force from how the gripper deforms ↓
Force is the missing signal in most robot learning data. Maybe we don’t need more electronics, wiring and latency to get it.
Huge thanks to Prof. Jianshu Zhou and Hao Wu at the NUS Grasping & Manipulation Lab.
Full write-up: https://t.co/3kPm5PzCOV
It can:
1/ track every marker in 97-99% of frames (clean background)
2/ estimate grip force within ~2.6 N with a 1-parameter physics model
3/ tell how different objects are held
The build is a Frankenstein of 2 open-source projects and a lot of plastic 😆
You don’t need to know how to code anymore. You just need an idea.
With Replit Agent, plain English becomes working software. And this weekend, you can experience it for FREE.
We’re hosting Replit's first Singapore Hackathon.
https://t.co/4TVutillJr
@Franciscocrz@Replit
What to expect:
• Founders sharing how they went from idea to live product fast
• Full access to Replit Agent at full capability, free for the weekend
• A short workshop + live demo to get you up and running at the beginning
https://t.co/4TVutillJr