Yichao Zhong (@PeterYChong) header
Yichao Zhong (@PeterYChong) profile picture

PhDing @HKUniversity Prev: Bachelor @sjtu1896; Internship @CMU_Robotics

Hong Kong SAR, China
Joined January 2024
Photo by Yichao Zhong (@PeterYChong)Photo by Yichao Zhong (@PeterYChong)Photo by Yichao Zhong (@PeterYChong)Photo by Yichao Zhong (@PeterYChong)Photo by Yichao Zhong (@PeterYChong)Photo by Yichao Zhong (@PeterYChong)Photo by Yichao Zhong (@PeterYChong)Photo by Yichao Zhong (@PeterYChong)Photo by Yichao Zhong (@PeterYChong)
The ⚽️ #WorldCup2026 kicks off tonight 🏆 — and we're sending a humanoid onto the pitch too! Meet RoboNaldo: a versatile humanoid football shooter. It can strike the ball up to 13m/s, accurately lands within 17 cm of a 3 m target away, outdoors on real grass pitch, off a live human pass, with fully onboard perception. opendrivelab.com/RoboNaldo 🧵👇
1
12
4
19
4,521
Yichao Zhong retweeted
Qiayuan (@qiayuanliao) and I wrote about modeling the Unitree G1’s closed-loop ankle linkage in @mujoco. We look at how the linkage changes the ankle’s motion and dynamics, and what it takes to simulate it reliably. We spent quite a bit of time on the blog post making interactive demos and shareable models, and we hope you find it useful! kzakka.com/closing-the-loop/
15
72
7
418
26,229
Yichao Zhong retweeted
Everyone's been showing Fable and GPT-6 Astra driving robot arms. We trained an open 27B Qwen model to do it, purely from its own attempts in simulation, with zero demonstrations. And it can run locally at 220+ tok/s on an RTX 5090. On a real robot it had never touched, it now succeeds in 18 of 30 tests, up from 0 before training. It also copies a painting and designs its own chemistry experiment. It generalizes and follows instructions better than π0.5, and on LIBERO it comes close to Astra max. All without losing its general capabilities, like math and coding. Built on Qwen3.8-27B. We've open-sourced the model weights, code, and full training infra! Check out our blog at mit-han-lab.github.io/pear-b….
54
170
52
1,254
117,869
Yichao Zhong retweeted
🚀 Introducing #compass1, an @Archon_Robotics initiative collaborated with our trusted partners, @UnitreeRobotics @HKUniversity @Huawei @boosterobotics @OpenDriveLab, etc. This navigation work, one-of-its-kind, complements the missing skillset for full-stack real world deployment from Archon model family. Kudos to the team! archon.tech/research/compass… – at Austin, TX
3
3
1
17
1,331
CoRL 2026 Oral 🌟 A humanoid robot can now kick a soccer ball at 47.2 km/h. Meet RoboNaldo, a new soccer-shooting framework from researchers at HKU, CUHK, and Archon Robotics. Deployed on a Unitree G1, it achieved sub-meter shooting accuracy from 3 meters away: • Stationary ball: 0.73 m average error. • Moving ball: 0.86 m average error. Peak post-contact ball velocity reached 13.10 m/s (47.2 km/h). That's 59–71% of reported professional open-play shot speeds. Previous approaches had two limitations: • Motion imitation: Following human kicking motions provides stability, but struggles to adapt to different ball positions and strike timings. • Pure reinforcement learning: Learning through trial and error makes it difficult to discover effective kicking motions from scratch. RoboNaldo combines both approaches in a three-stage motion-guided RL framework. Stage 1: Learn a stable kick. Use a human kicking motion as a reference to learn coordinated, stable movements. Stage 2: Adapt to different ball positions. Randomize ball positions and use RL to fine-tune the kicking motion for different shots. Stage 3: Shoot moving balls. Combine locomotion control with kick timing, allowing the robot to approach moving balls and strike at the right moment. This takes RoboNaldo beyond stationary shots, enabling one-touch shots on moving balls, much closer to real soccer scenarios.
The ⚽️ #WorldCup2026 kicks off tonight 🏆 — and we're sending a humanoid onto the pitch too! Meet RoboNaldo: a versatile humanoid football shooter. It can strike the ball up to 13m/s, accurately lands within 17 cm of a 3 m target away, outdoors on real grass pitch, off a live human pass, with fully onboard perception. opendrivelab.com/RoboNaldo 🧵👇
2
1
9
740
RoboNaldo is heading #CoRL2026, selected as Oral. Oral: Nov 10 11:30-12 AM | Poster: #181 Nov 10 5-6:30 PM Sad that I cannot attend in person. My advisor @francislee2020 will present the oral. Come stop-by and talk to him and my other colleagues there!
2
14
700
Yichao Zhong retweeted
With @JayHe748646 @KehanWen170077 @leggedrobotics , we push humanoid mobility to next frontier. Achieved by a generalist contact planner + a generalist tracker + a generalized neural mapping, warmstarted from few human data. Website: leggedrobotics.github.io/ACP… Promo Video 👇
27
75
21
528
55,412
Yichao Zhong retweeted
We found a way to scale humanoid control that reduces GPU-hours by more than 1000×. Across 129,514 motions, SuperMimic surpasses the state of the art (128 GPUs, 7 days) with 1 GPU in 3 hours. Project: supermimic.github.io Demo: supermimic.github.io/demo/
98
319
124
1,925
292,571
Yichao Zhong retweeted
Not even a surprise.
following the discussion weeks ago, I asked codex to vibe a comparison for MPC vs RL for just table-top EE tracking, 1.6Mtok gone. It uses Pinocchio for MPC, and codes a cupy-based simulator, tuned RL to achieve better tracking accuracy (actually both errors are too small)
1
3
60
6,807
Your cerebellum is much smarter than you expect, so should robots be! A bare Motion Tracker's never gonna be the goat S0
My blog on System 0 in the Astra era: As spatial grounding (S1/2) improves, we need dynamics-grounded controllers (S0) and expressive interfaces that let Astra steer physical interaction. Hope this resonates or sparks disagreement. I’d love your thoughts! haoyangweng.pages.dev/blog/2…
1
7
1,233
Never had multi humanoids working so close…🫣
What’s better than a humanoid playing ping-pong? Two playing doubles—and trying not to crash into each other 😁 Meet PingpongDuo: humanoid table tennis doubles 🏓 Check out the project and play with interactive 3D scenes archon.tech/research/Pingpon… Paper: arxiv.org/abs/2610.20579
105
To make a locomotion policy adapted to multiple kinds of shoes 🥲
Humans expand where they can walk by changing shoes. Can robots do the same? Put shoes on a quadruped and its barefoot policy degrades. SHOES recovers it with 10 minutes of RL on the real robot 🥾 CoRL 2026🎉 [1/9]
2
109
🤖 Building Whole-Body Intelligence for the Real World. Were Just Getting Started. From bending down to pick up objects and navigating steps to autonomously clearing tables, operating dishwashers, and organizing refrigerators, our robots are learning to perceive, move, manipulate, and coordinate their entire bodies in real-world environments. Powered by our Whole-Body Intelligence (WBI) framework, were building generalist humanoid foundation models by combining large-scale human whole-body data pretraining, context-aware behavioral foundation models, and efficient post-training with real-robot data - turning human experience into real-world robotic skills. 👉 Watch Whole-Body Intelligence in action in our three-minute demo. More to come. Stay tuned at archon.tech #ArchonRobotics #WholeBodyIntelligence #WBI #Humanoid #PhysicalAGI
22
43
9
306
62,626
Can't agree more with the opinion acknowledging onboard perceptions. In my CoRL'26 work on accurate robot soccer shooting (opendrivelab.com/RoboNaldo), I used onboard LiDAR+D435I camera for perceptions. I can say >90% of my effort was put on making good ball perception. Robust perception(and also its connections to robot) on a wildly dynamic robot is truly hard. Camera image blurs a lot🥲 Also, when your perception is bad, estimation comes important. We put Kalman Filter here; and MIT's yoga has a learned estimator from real-world data, that's our weapon to work with poor sensors.🥲
What do robot sports teach us? We have seen incredibly cool progress in robot soccer and tennis recently, but just as interesting is what we definitely do not see. Robots are increasingly capable of incredibly fluid, reactive whole body control - but their ability to perceive the world while doing so is incredibly limited. And so we see shortcuts being taken, external trackers used, to avoid the limitations of onboard sensing and sensorimotor control while in challenging environments. The future of robotics will require closing this gap. Some thoughts -> itcanthink.substack.com/p/wh…
9
371
What a precise strike🤩
the highlight of IROS
103
Accurate striking while jumping, impressive. It’s hard for some human
Can a humanoid be your badminton buddy, learning from just 30 min of human motion data? Humanoid Badminton 🏸, accepted at #CoRL2026, learns diverse racket skills with a single policy, enabling natural whole-body motion and real-world multi-skill rallies with forehand, backhand, and highly dynamic jump returns. 🌐 Project: sunlight02.github.io/humanoi…
9
285
Yichao Zhong retweeted
As robot learning leans more and more toward scaling, it can seem like there is less and less left for academic labs to do. I think humanoid ball sports is a topic especially well suited to academia: it doesn't need huge amounts of compute or data, but it does take a lot of effort and ingenuity. And it's moving fast: my timeline has been full of humanoid robots playing ball lately. So I made Awesome Humanoid Ball Sports 🏓🎾🏸⚽🏀, a curated list of research on humanoid table tennis, tennis, badminton, soccer and basketball, to make literature review easier and help everyone keep track of the latest work. It also comes with an interactive map of which groups work on which sport.
2
4
1
26
2,511
With huge respects, excellent work! Also check out our recent open-sourced work on humanoid soccer shooting, lidar+vision-based, mocap-free, in the wild: Paper: arxiv.org/abs/2606.11092 Page: opendrivelab.com/RoboNaldo/ Code: github.com/opendrivelab/Robo… Feels like the sporting humanoids era🥲
Introducing Joga, a full-stack humanoid soccer system with active vision. Agile dribbling. Tight direction changes. Receive → dribble → shoot/pass. An actuated neck enables ball tracking with onboard vision. Residual models for perception + actuation close the sim-to-real gap. Paper coming soon! Work done in collaboration with @Avi_Narula14 @khai_ngx @venki1899 @JohnMarangola @martinpeticco @pulkitology
4
178
🚨Thrilled to announce: ⚽️RoboNaldo is accepted by #CoRL2026. See you guys in Austin!!!
The ⚽️ #WorldCup2026 kicks off tonight 🏆 — and we're sending a humanoid onto the pitch too! Meet RoboNaldo: a versatile humanoid football shooter. It can strike the ball up to 13m/s, accurately lands within 17 cm of a 3 m target away, outdoors on real grass pitch, off a live human pass, with fully onboard perception. opendrivelab.com/RoboNaldo 🧵👇
1
14
491