Ph. D. candidate @ UT Dallas. Member of the Intelligent Robots and Vision Lab (IRVL)
Dallas, Texas
Joined December 2017
- Tweets299
- Following139
- Followers391
- Likes3.7K
Pinned Tweet
We’re excited to share our work on cross-embodiment dexterous manipulation:
The Unified Hand Action Space (UHAS) — a new representation that allows a single policy to control robotic hands with completely different kinematic structures and numbers of fingers.
🎥 This video shows one policy simultaneously controlling four different hands.
Project: irvlutd.github.io/UHAS/
🧵 Thread
1/9 🤖
Luis Felipe Casas retweeted
The odd genius who made this must now step forward and accept the crown as the most brilliant living thing
Luis Felipe Casas retweeted
For students working on #robotics: Make sure you understand coordinate frames, SE(3) transformations, and robot kinematics.
Robot learning is not just about training models on data. Robot data has physical meaning, and understanding the underlying geometry is essential.
Luis Felipe Casas retweeted
Finished fine-tuning MolmoAct2 on our VLA-Replica SO-101 setup, using 50 demonstrations per task!
One result I find interesting: it outperforms π₀.₅ on OOD by 10%.
Check out our leaderboard:
irvlutd.github.io/VLAReplica… teapaw.org/maylynn_e/status/21002…
Luis Felipe Casas retweeted
We recently fine-tuned Nvidia's GR00T N1.7 VLA model w/ 50 demos, and evaluated across our benchmark, VLA-Replica.
It performs relatively similar to π₀ in ID settings, and π₀.₅ during OOD scenes.
Check out our leaderboard: irvlutd.github.io/VLAReplica…
We built a live visualization for UHAS — so you can see how one shared action maps to completely different hands. Drag a driving plane on the sphere and Allegro, LEAP, Shadow, MANO, and Wuji all move at once. Same deformation. Different kinematics. Coordinated poses. The policy never outputs joints. It deforms a canonical sphere. Our method turns that geometry into each hand in real time.
#DexterousManipulation #Robotics #EmbodiedAI 🤖
We’re excited to share our work on cross-embodiment dexterous manipulation:
The Unified Hand Action Space (UHAS) — a new representation that allows a single policy to control robotic hands with completely different kinematic structures and numbers of fingers.
🎥 This video shows one policy simultaneously controlling four different hands.
Project: irvlutd.github.io/UHAS/
🧵 Thread
1/9 🤖
Huge thank you to the @personaaiinc team for an incredible Summer internship.
The learning experience was massive and the people are truly awesome. Grateful for everything — and excited to stay connected as you help build an amazing future. Wishing the whole team continued success!
#PersonaAI #Internship #Humanoids
Luis Felipe Casas retweeted
How can robots handle dynamic environments when their demonstrations only show static objects?
Introducing SIDO, a framework that bridges this gap by turning static-only demonstrations into policies that can track moving targets, without requiring additional data collection.
Huge thanks to @chris_j_paxton and @DJiafei for having us on @RoboPapers!
It was a pleasure discussing UHAS and the challenges of representing actions across different robot hands. Excited to share more about unified action spaces for cross-embodiment dexterous manipulation.
Watch Episode #97 and check out the project: irvlutd.github.io/UHAS/
Representing actions for robot hands is actually very challenging. Robot hands have different sizes and shapes, and different degrees of freedom. Representing their actions purely as joint states — as we naively would — leads to less transfer across the wide variety of available hands. Instead, could we design a universal hand action space to allow robots to learn cross-embodiment dexterous manipulation more easily?
@lfcasas7 and @YuXiang_IRVL come on to tell us about UHAS: the Unified Hand Action Space, a sphere-based action representation which enables cross-embodiment learning of dexterous manipulation skills.
Watch Episode #97 of RoboPapers, with @chris_j_paxton and @DJiafei, now to learn more!
Luis Felipe Casas retweeted
Representing actions for robot hands is actually very challenging. Robot hands have different sizes and shapes, and different degrees of freedom. Representing their actions purely as joint states — as we naively would — leads to less transfer across the wide variety of available hands. Instead, could we design a universal hand action space to allow robots to learn cross-embodiment dexterous manipulation more easily?
@lfcasas7 and @YuXiang_IRVL come on to tell us about UHAS: the Unified Hand Action Space, a sphere-based action representation which enables cross-embodiment learning of dexterous manipulation skills.
Watch Episode #97 of RoboPapers, with @chris_j_paxton and @DJiafei, now to learn more!
Luis Felipe Casas retweeted
Full episode dropping soon!
Geeking out with @lfcasas7 @YuXiang_IRVL on Cross-Embodiment Robot Manipulation via a Unified Hand Action Space irvlutd.github.io/UHAS/
Co-hosted by @chris_j_paxton @DJiafei
Luis Felipe Casas retweeted
This action representation isn't limited to cubes. It can in-hand manipulate some other objects too
We’re excited to share our work on cross-embodiment dexterous manipulation:
The Unified Hand Action Space (UHAS) — a new representation that allows a single policy to control robotic hands with completely different kinematic structures and numbers of fingers.
🎥 This video shows one policy simultaneously controlling four different hands.
Project: irvlutd.github.io/UHAS/
🧵 Thread
1/9 🤖
Luis Felipe Casas retweeted
We’re excited to share our work on cross-embodiment dexterous manipulation:
The Unified Hand Action Space (UHAS) — a new representation that allows a single policy to control robotic hands with completely different kinematic structures and numbers of fingers.
🎥 This video shows one policy simultaneously controlling four different hands.
Project: irvlutd.github.io/UHAS/
🧵 Thread
1/9 🤖
Luis Felipe Casas retweeted
I'll be presenting this work at the 4th Workshop on Dexterous Manipulation this morning!
Feel free to stop by our poster and chat.
dex-manipulation.github.io/r…
We’re excited to share our work on cross-embodiment dexterous manipulation:
The Unified Hand Action Space (UHAS) — a new representation that allows a single policy to control robotic hands with completely different kinematic structures and numbers of fingers.
🎥 This video shows one policy simultaneously controlling four different hands.
Project: irvlutd.github.io/UHAS/
🧵 Thread
1/9 🤖
Luis Felipe Casas retweeted
🎊 Egoengine will appear as Spotlight in #RSS2026 4th Workshop on Dexterous Manipulation. Tune in on July 13th 9:50-10:20am 🇦🇺.
We present EgoEngine, a scalable framework for dexterous manipulation learning. It converts egocentric human videos into robot executable demonstrations through motion retargeting, RL refinement, and large scale simulation. Dive deeper in @Randle_Liu’s thread.
Luis Felipe Casas retweeted
A humanoidal welder! 👨🏼🏭
@personaaiinc deployed their humanoid to a real welding facility in Houston and executed an actual weld.
All of that in a hot-work industrial environment with arc flash, smoke, and uneven flooring.
The robot squats down, executes the weld with a steady arc, stands back up.
Functional welding requires multi-joint precision, spatial awareness, and real-time path correction.
The robot had to hold a lit torch, maintain arc stability, and lay down a valid weld path.
The motion is choppy. They're honest about that. But they proved the robot can handle the full sequence that actual industrial work demands.
Get low, do precision work, stand up, move to the next spot. That's the rhythm of real welding jobs.
Teleoperation is their training method. VR headset interface captures expert demonstrations.
Those become the data to train neural networks toward eventual autonomy.
This is what deploying into real facilities means.
That's a nice one gents @nicolausradford!
Here's more detailed blog: persona.ai/robot-welding-in-…
~~
♻️ Join the weekly robotics newsletter, and never miss any news → ziegler.substack.com