Using a tray as the robot’s end-effector opens the door to:
✅ Moving multiple objects at once
✅ Handling large, fragile, or oddly-shaped items
✅ Lower cost (no gripper = fewer moving parts)
🧵2/7
Proud to be part of this small but mighty group - let me know if I missed you or any women to add!
go.bsky.app/BuDHCYF
🎉Excited to share that our paper was a finalist for best paper at #HRI2025! We introduce MOE-Hair, a soft robot system for hair care 💇🏻💆🏼 that uses mechanical compliance and visual force sensing for safe, comfortable interaction. Check our work: moehair.github.io @cmurobotics.bsky.social 🧵1/7
Annika Thomas
Yuemin Mao
Uksang Yoo
📊 Results:
We test the method on a UR5e robot with 12 different object configurations and compare it to the standard Coulomb friction model.
🟢 Our method reduces object displacement by up to 86% 🎉
🧵6/7
Yuemin Mao
📄 “Hearing the Slide: Acoustic-Guided Constraint Learning for Fast Non-Prehensile Transport”
By: @yuemin-mao.bsky.social, @bardienus.bsky.social, Moonyoung Lee, @jeff-ichnowski.bsky.social
arXiv: arxiv.org/abs/2506.09169
🎥 Website + videos: fast-non-prehensile.github.io
🧵7/7
We train a friction model that maps:
object slipping → tray speed + acceleration → real-world “friction coefficient” .
This becomes a dynamic constraint in a time-optimized motion planner. Now the robot knows how fast it can move without losing the object 🤖。
🧵5/7
Most planners rely on the standard Coulomb friction model to prevent object sliding.
That works in theory…But in practice 🤔? At high speeds, robot vibrations and subtle dynamics cause objects to move well before the model predicts 🫨.
🧵3/7
Sliding makes noise 🔊. When an object slips, it vibrates the tray—and we can hear it 👂. By attaching a contact mic 🎙️, we capture these signals at high frequency, low latency, and low cost. From acoustic data, we learn how robot speed and acceleration affect friction ✔️
🧵4/7
Yuemin Mao
Yuemin Mao
Yuemin Mao
Yuemin Mao
🤖📦 Want to move many items FAST with your robot? Use a tray. But at high speeds, objects may fall off 💥.
Introducing our new method: it hears sliding 🎧, learns dynamic friction 🥌, and plans time-optimized motions to transport objects 🚀.
fast-non-prehensile.github.io/
🧵1/7
Imagine if robots could fill in the blanks in cluttered scenes.
✨ Enter RaySt3R: a single masked RGB-D image in, complete 3D out.
It infers depth, object masks, and confidence for novel views, and merges the predictions into a single point cloud. rayst3r.github.io
