
Doctoral Student at @crl_ethz @ETH, working on loco-manipulation, robot learning with @StelianCoros and @GuanyaShi. Cells interlinked within cells, interlinked.
Zurich, Switzerland
Joined January 2018
- Tweets276
- Following875
- Followers893
- Likes2.6K
Happy to share KINO, a keyframe-based interface between VLM and humanoid low-level controller! Kudos to @kkkakawe for the amazing project. #ai #humanoid #rl #vlm
How should a vision-language model talk to a humanoid's whole-body controller? 🤖
Our new paper, KINO, proposes a simple answer: motion keyframes.
📄 arxiv.org/abs/2609.18869
🎥 sc0k.github.io/kino/
🎞️👇
🤖 Two papers at #CoRL2026 in Austin on Nov 9!
🐕 Teaching Robots Like Dogs — Taerim Yoon | Poster #165, 5–6:30 PM
⚡ TARC — Arnav Sukhija | Spotlight 2:30–3 PM + Poster #62, 5–6:30 PM
I can't attend in person, but please stop by and meet Taerim & Arnav! 🚀
🤖 Made with AI
🐕 Teaching Robots Like Dogs
We teach quadruped robots agile navigation skills through luring, gestures, and speech, achieving 97% task success with less than 1 hour of human demonstrations!
📄 Paper: arxiv.org/abs/2601.08422
🎤 Presented by Taerim Yoon @TaerimYoon_ at #CoRL2026
1
1
128
⚡ TARC: Time-Adaptive Robotic Control
Instead of controlling robots at a fixed frequency, we let them learn when to act, matching high-frequency control performance with less than half the control frequency!
📄 Paper: arxiv.org/abs/2510.23176
🎤 Presented by Arnav Sukhija at #CoRL2026
1
97
(Maybe) the first deployable humanoid tracking policy trained via differentiable simulation! Happy to share our recent work on using bundled gradient to help stablizing the gradient information for stiff contact scenarios. Congrats! @dyumanaditya
#robotics #rl #diffsim #humanoid
🤖 Ever wondered if you can train end-to-end humanoid policies entirely within a differentiable simulator?
🥁 Introducing 𝗕𝘂𝗻𝗱𝗹𝗲𝗱 𝗖𝗼𝗻𝘁𝗮𝗰𝘁 𝗚𝗿𝗮𝗱𝗶𝗲𝗻𝘁𝘀 (𝗕𝗖𝗚)
🌐: bundledcontactgradients.gith…
📄: arxiv.org/abs/2609.30951
🧵👇
Carlo is incredible! Welcome back!!
Incredibly excited to return to ETH Zurich as Assistant Professor of Robotics and Artificial Intelligence next summer. I'll be recruiting PhD students to build a new lab on general-purpose embodied intelligence for humanoid robots and dexterous manipulation. Details coming soon!
1
6
898
LURE has been accepted to CoRL!
Introducing our recent work: 🐶Teaching Robots Like Dogs: Learning Agile Navigation from Luring, Gesture, and Speech
We ask a simple question: Can we teach robots the way we teach dogs?
Project page: lnkd.in/eWWSJds2
📄 2 papers got accepted to CoRL this year @corl_conf! Huge congrats to the awesome collaborator @TaerimYoon_ and my awesome student Arnav Sukhija! 🦾🦾
TARC: arxiv.org/abs/2510.23176
LURE: arxiv.org/abs/2601.08422
Jin Cheng retweeted
Behavioral cloning mystery
seohong.me/blog/behavioral-c…
I wrote a new blog post about "mysteries" in behavioral cloning that appear with real-world robot data (e.g., overfitting is "good"). I also tried to demystify them and shared my thoughts!
📄 New paper alert! We are happy to introduce FB-MEBE, a behavior foundation model (BFM) for quadrupeds WITHOUT precollected dataset! The paper has been accepted to @RL_Conference. Kudos to @jhu486 and @nnuriarmengol! Please checkout our website: tinyurl.com/fbmebe
Excited to introduce FB-MEBE, led by @jhu486 ! A Behavior Foundation Model (BFM)🤖 for quadrupedal control trained without rewards AND without pre-collected data! Given a reward at test-time, it recovers an optimal policy zero-shot and transfers seamlessly to real hardware.🧵
Jin Cheng retweeted
Monday thoughts:
Since deep learning, there are two “root-level” paradigm shifts in robot learning: Sim2Real for locomotion and Behavior Cloning (BC) for manipulation. 1/
Jin Cheng retweeted
Can learned deformable manipulation policies improve at inference time — without retraining?
Yes: with real2sim2real + sampling-based MPC.
🟣 Novel physics-grounded deformable simulation with fast, accurate, GPU-parallel real-time rollouts
🟣 RGB-based real2sim state estimation
🟣 Policy-agnostic refinement that boosts inference-time success rates
🟣 Runs on a single RTX 4090 GPU
Project page: silr-cloth.github.io/
More details in the thread below 👇
Jin Cheng retweeted
Humanoid robots don't need to look human.
Meet Eno, our first general-purpose robot.
Not a machine pretending to be human, but intelligence given a body.
At Genesis, we’re building a future where robots don’t feel cold or distant, but capable, calm, and ready to help.
Available Q4 this year.
Jin Cheng retweeted
🐊Tomorrow, our student Yuanchen Yuan is presenting CAIMAN, “Causal Action Influence Detection for Sample-efficient Loco-Manipulation” at @ieee_ras_icra.
Proud to have contributed alongside Yuanchen, Jin Cheng @catachii and Stelian Coros.
Go talk to them ⭐️
Replying to @crl_ethz
📅 WEDNESDAY
⏰ 09:00 - 10:30 | Poster WeI1I.67
📄 CAIMAN: Causal Action Influence Detection for Sample-Efficient Loco-Manipulation
👥 Yuanchen Yuan, Jin Cheng @catachiii , Nuria Armengol Urpí @NriaArmengol2 , Stelian Coros @StelianCoros






















