
Actively looking for Summer Internship 2027 outside the US! CS PhD at Uchicago @HaozhiQ | Prev at Georgia Tech @danfei_xu | Robotics, Computer Vision
Chicago, IL
Joined November 2022
- Tweets125
- Following609
- Followers397
- Likes1.9K
Yangcen Liu retweeted
What should transfer from a human video to a robot?
Human2Any learns how objects interact, then adapts that knowledge through test-time reasoning and planning.
Accepted to #CoRL2026! I'll be presenting in person next week—come say hi!
human2any.github.io (1/n)
📢 Call for Papers!
Excited to serve as a Guest Editor for the IEEE Transactions on Haptics Special Issue on The Sense of Touch in Contact-Rich Dexterous Robot Manipulation!
We welcome work on tactile sensing, haptic teleoperation, robot learning, and contact-rich manipulation.
📅 Deadline: March 15, 2027
shorturl.at/QXBvR
together with @wenzhenyuan382, Rebecca Friesen, Yasemin Bekiroglu, @yushe_1, and @linshaonju.
Come and chat with us @alfie14x @XinchenYinYXC @woochulshin1726 @ShuoCheng94 for EgoEngine! (Guess who is not invited 😅) Also I am looking for 2027 summer internship outside the US.
Heading to #CoRL2026 next week! Excited to share 4⃣ papers during the poster sessions.
I will also present EgoEngine, which was selected for a Spotlight Oral Presentation at the BiDexTool Workshop! 🎉.
If you’re attending and working on robotics, foundation models, reinforcement learning, or agentic AI, let’s connect! Happy to grab a coffee and chat. ☕
I’m also exploring full-time opportunities for 2027. If your team is hiring or you know of any interesting opportunities, feel free to reach out!
alfred-cueva.github.io/
Egoengine: egoengine.github.io/
Human2Any: human2any.github.io/
WT-UMI: wt-umi.github.io/WTUMI/
SIDO: sido-staticindynamicout.gith…
Yangcen Liu retweeted
Heading to #CoRL2026 next week! Excited to share 4⃣ papers during the poster sessions.
I will also present EgoEngine, which was selected for a Spotlight Oral Presentation at the BiDexTool Workshop! 🎉.
If you’re attending and working on robotics, foundation models, reinforcement learning, or agentic AI, let’s connect! Happy to grab a coffee and chat. ☕
I’m also exploring full-time opportunities for 2027. If your team is hiring or you know of any interesting opportunities, feel free to reach out!
alfred-cueva.github.io/
Egoengine: egoengine.github.io/
Human2Any: human2any.github.io/
WT-UMI: wt-umi.github.io/WTUMI/
SIDO: sido-staticindynamicout.gith…
I am looking for Summer 2027 internship opportunities outside the US!!! I am working in agentic robotics, dexterous manipulation, and learning from human videos. Here is my homepage: yangcenliu.github.io/. Please reach out to me if there is any opportunities!
3
46
11,678
We have seen quite a few capable agents controlling robots. But an important next question is: how can we build an automatic pipeline for these agents to keep improving themselves?
Check out RPG.
Starting from an offline robot dataset, RPG automatically figures out what capabilities need to be practiced, builds sim environments for those skills (practice tasks, rather than digital twins), and then improves the robot’s skill library through repeated rounds of practice and failure analysis.
Then, the improved system also transfers back to the real world, achieving 30/30 successful hardware trials across three tasks.
Today we use datasets mainly to train robots. In the future, datasets can also tell robots what they should practice next. As robot data becomes increasingly abundant, automatically turning those experiences into targeted practice and continual self-improvement feels is really promising!
Practice makes perfect. A truly agentic robot shouldn’t just know what to do. It should also figure out how to get better.
What if a robot could watch a demonstration, figure out what it needs to practice, and learn how to get better through practice?
Introducing RPG: Guided Self-Improvement for Embodied Agents, a real-to-sim-to-real pipeline that turns real-world demonstrations into targeted practice.
RPG reconstructs practice tasks from demonstrations, uses demo videos to guide skill improvement, and continually updates a shared skill system from failures.
On real robots, RPG achieves 30/30 successes across 3 tasks. In simulation, it reaches 95.0% success across 22 manipulation tasks after 15 rounds of self-improvement.
Demo → Reconstruct → Guided Practice → Self-Improve → Go Real
Yangcen Liu retweeted
One human demonstration. Any multi-fingered hand. Zero-shot sim-to-real visuomotor policy.
morphometricimitation.github…
Collaborators: @he_siming @ckwolfeofficial @HaozhiQ @LeaMue27 Shankar Sastry, Claire Tomlin, @JitendraMalikCV
A near future direction like V2S2R: powerful large language models take high level reasoning, policy for low-level control. Code-as-Policy is getting viral, but more work will focus on replacing code with atomic skill policies for better generalizability and flexibility.
After GPT6 helps you create the simulation, how can you turn it real world skills?
See our works below for insights!
1
9
1,378
Yangcen Liu retweeted
Every new robot hand claims to be the most dexterous. But there's never an objective, useful number behind it.
KaRMA is the first metric to compute a hand's true dexterity using just a kinematic model. (IROS 2026) (1/9)
Our paper EgoEngine got accepted to CoRL 2026 🎉🎉🎉 Thanks again to our great team! See you in Austin! @ShuoCheng94 @XinchenYinYXC @alfie14x @woochulshin1726 !
The web is full of egocentric human videos. But robots can’t directly use them as demonstrations yet.
Meet EgoEngine: From Egocentric Human Videos to High-Fidelity Dexterous Robot Demonstrations, for zero-shot visuomotor learning without real-robot demos.
egoengine.github.io
Very cool! To optimize the performance we should focus more on morphology besides policy considering robotics is an integrated system problem.
Transformer Transformer will be presented at #CoRL2026 🚀 See y'all in Austin!
Website: transformer-transformer.gith…
Video: youtube.com/watch?v=TTyjvPVF…
Paper: transformer-transformer.gith…
14
1,855
Yangcen Liu retweeted
Coming back to Shenzhen! Who can invite me to your robot learning startups / big tech? @Randle_Liu and I would love to grab coffee and chat together👀
Yangcen Liu retweeted
Train on static objects. Grasp moving objects. 🤖
Many visuomotor policies assume static objects. Real deployments break this: parts drift on conveyors, fruits sway in the wind. Meet SIDO (Static In, Dynamic Out), taking a policy trained only on static demos to moving objects.
Exciting experience in WRC26. Unfortunately no big change compared to last year. Only vertical scenes with VA (No L) works and still a long way to convergence. Investment is cooling down and it’s our duty to push for the next revolutionary technical breakthrough.
1
8
1,348
I’ve seen how torturing and challenging WARP is doing started from a single trajectory replay to ultimate policy. But glad to see @DanielZhenyang and the team made it! Another giant step to use human videos as a source for policy training!
Human data is the cheapest source of whole-body mobile manipulation demos. Turning it into robot data is the hard part.🤖
WARP retargets OFFLINE human motion into replayable robot actions in closed form — precise, consistent, faithful to whole-body intent.
🌍 warp-retargeting.github.io/
That’s a real reflection. I was wondering why Tesla and Figure (1X recently) the leading producers are using tendon-driven hands as a fragile and delicate art pieces for massive heavy dirty work while others just use direct driven hands or cheap grippers following 20:80 rule.
The biggest engineering mistake I made at Figure was building a tendon-based hand
Our first hand design in 2022 was a tendon hand for our F.01 robot. At a high level, the tendon approach sounds appealing, which is why I chose it: you get more space for packaging actuators since the forearm is larger, potentially higher degrees of freedom, and it's biologically inspired
We built and manufactured this entire hand and tested it in early 2023. If you saw it in person, it was truly an engineering work of art. It turned out to be one of the worst engineering decisions I've made, maybe the worst in four years. Tendons are a complete local maximum, and that only becomes clear in hindsight, after exploring every other possible hand design. Figure is unique in that we've now built and tested several hand architectures, which gave us a clear sense of where to head
Figure is working on our next generation hand, and it's so much better than anything you can do with tendons, it's not even funny
Human video pretraining is proved to be vital!
Today we are introducing Dyna-2, a world-action model pre-trained on one million hours of human video. At this scale, for the first time, we discovered several new scaling laws:
• world-action models exhibit scaling law on human data across four orders of magnitude, from 1000 to 1,000,000 hours,
• this human data scaling law implied a scaling law on never seen robot data,
• both data and objective matter; world modeling and scaling on video data are essential for cross-embodiment scaling transfer to emerge
🧵
10
994
How to efficiently deploy multiple robots simultaneously on large policy models like VLA is a valuable and hard problem that researchers these days did not pay enough attention: check @davidhe137 and @rbansal22’s wonderful problem formulation and solution.
🤖 Scaling deployment in robotics can be difficult and expensive. How can we properly serve a VLA/WAM to 10+ robots concurrently from a single GPU?
Introducing Armory, a system for serving large robot policies to a fleet of robots from the cloud/edge. Through real-time scheduling of action chunks, Armory boosts task throughput over baselines by up to 18% on heterogeneous fleets in the real world.
🧵1/6























