Ziqi Han

Tongji University · RoboParty Lab | Humanoid robotics, robot–object interaction & unsupervised RL

[Qiayuan Liao @qiayuanliao]

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: https://t.co/6gc9WKm9qY
Demo: https://t.co/OHRvx5ogQY

[Yitang Li @li_yitang]

Can humanoids adapt to new terrain using only flat-ground references? 🤖

Meet P-BFM! 🥳

We learn a terrain-agnostic intent space with unsupervised RL and forward-backward representations, paired with a terrain-aware policy that adapts across diverse terrains—without terrain-specific demonstrations.

📄🎬💻 Check out our paper, video, and code here!

🌐 https://t.co/0c7EW7x6Hu

[sudarshan s. harithas @sudarshan_s_h]

Excited to share 𝗕𝗥𝗔𝗖𝗘: a whole-body tracker for humanoids that can exert forces from diverse poses and adapt to varied terrains.

Humanoid whole-body trackers are typically trained to follow references recorded on flat ground, carrying nothing. As a result, they know 𝗻𝗲𝗶𝘁𝗵𝗲𝗿 𝘁𝗵𝗲 𝗳𝗼𝗿𝗰𝗲𝘀 𝘁𝗵𝗲 𝗿𝗼𝗯𝗼𝘁 𝗺𝘂𝘀𝘁 𝗲𝘅𝗰𝗵𝗮𝗻𝗴𝗲 𝘄𝗶𝘁𝗵 𝗼𝗯𝗷𝗲𝗰𝘁𝘀 𝗻𝗼𝗿 𝗰𝗮𝗻 𝘁𝗵𝗲𝘆 𝗮𝗱𝗮𝗽𝘁 𝘁𝗼 𝗱𝗶𝘃𝗲𝗿𝘀𝗲 𝘁𝗲𝗿𝗿𝗮𝗶𝗻𝘀. BRACE closes both gaps in a single tracker.

Our key innovation is a 𝗿𝗲𝗳𝗲𝗿𝗲𝗻𝗰𝗲-𝗮𝗱𝗮𝗽𝘁𝗮𝘁𝗶𝗼𝗻 𝗳𝗼𝗿𝗺𝘂𝗹𝗮𝘁𝗶𝗼𝗻. Given humanoid reference trajectories, we adapt them to the terrain and the commanded force, then train the controller on these adapted references. Tracking them naturally produces the stance and bracing needed to handle the terrain and exert the force (or compensate for external loads). 𝗧𝗵𝗲 𝗿𝗲𝘀𝘂𝗹𝘁𝗶𝗻𝗴 𝗰𝗼𝗻𝘁𝗿𝗼𝗹𝗹𝗲𝗿 𝗱𝗲𝗽𝗹��𝘆𝘀 𝗱𝗶𝗿𝗲𝗰𝘁𝗹𝘆 𝗼𝗻 𝘁𝗵𝗲 𝗿𝗲𝗮𝗹 𝗿𝗼𝗯𝗼𝘁.

To know more, please visit our:
Project page: https://t.co/YXmt6GZoFp
Interactive demo: https://t.co/UP6CDAKr8C
More details in this thread🧵
1/n

[RoboParty @RoboParty_DEV]

🚀RP1 is officially here.

🌍Putting control of robotics back in developers’ hands.
Introducing RP1 (ROBOTO 01) by RoboParty—a high-performance, fully open-source humanoid robotics platform for frontier robotics research and embodied AI model development.

🤖 Open across the full hardware and software stack:

• Robot platform: mechanical CAD, bill of materials, and assembly documentation
• Actuators: in-house ROMOMO joint designs and related technical resources
• Control: electrical systems, firmware, low-level drivers, and motion control implementations
• Development: simulation models, SDKs, data processing tools, and training tools
• Engineering foundations: technical documentation and research materials

Open the black box.
Understand the system.
Make it your own.

A deviation in motion can originate in the model, control policy, actuator response, or mechanical structure.
RP1 opens up the design and implementation of these layers so developers can trace problems, modify the system, and validate changes on real hardware.

Open-source algorithms for advanced motion control
RP1 has demonstrated continuous walking, recovery from disturbances, getting back up after a fall, and human motion tracking.

UFO and MimicLite connect motion learning, control policies, and execution on real hardware.

The open toolchain connects data collection, simulation training, hardware deployment, and experimental validation—supporting joint iteration across models, control, and hardware.

From RPO to RP1, we’re turning one team’s engineering experience into a starting point for many more.

Opening the future of embodied AI to everyone ready to explore it.

🔗RP1 on GitHub:
https://t.co/kMqFLi3eJ3
#RoboParty #RP1 #OpenSourceRobotics #EmbodiedAI

[Octi Zhang @octi_zhang]

(1/5) Everyone says dexterous manipulation needs real-world data. These policies were trained with none. Our robots mesh gears, thread nuts, and insert pegs zero-shot from RGB, and the same recipe works for locomotion across terrains much harder than before.

The missing piece wasn't a better RL algorithm, but a small tweak to what the robot practices.

Today we're releasing Success-Guided Sampling (SGS), a simple change to the task sampler that unlocks new levels of dexterity and agility with large-scale, end-to-end sim2real RL.

No demonstrations, no per-task reward tuning. Just PPO at scale.

Check out the policies yourself at https://t.co/dnwUr0tabh! All videos at 1x speed! 🧵 🔈

[RoboParty @RoboParty_DEV]

🤖 Coming soon. 🚀

RP1 — The fully open-source humanoid robot.

Built for developers. Born for intelligence.

#RoboParty #RP1 #OpenSource

[Ziqi Han @HanZiqi666]

Proud to be a co-author of I-BFM! What stayed with me was watching the real G1 recover, return to the box, and finish the carry. LOGO helps one shared policy make the next contact decision while keeping the task goal in view. Grateful to the whole team! 🤖📦

[Junhan Sun @DavidSunok47]

JEPAs & BFMs folks, Remember INTACT in July?

From JEPA into RL: meet LOGO in I-BFM 🤖
Local intent. Long-horizon purpose. Two horizons, one actor.

G1 MuJoCo carry success: 51.0% → 94.3% with LOGO.
🎬 No retraining. Just use next reward, next action.

How it works + paper & code 👇

[Ziqi Han @HanZiqi666]

5/5 Real G1: carry, push, and kick demos. Carry trials show a retry after failed box handling and recovery from a disturbance. Experiments use a 0.7 kg box, 50 Hz control, and motion-capture feedback. Next: onboard sensing, more objects, and planning.

[Ziqi Han @HanZiqi666]

Can humanoids recover in interaction tasks? 📦🤖
Excited to share I-BFM! Reward-conditioned policy: carry, push, kick. FB: robot–object–contact dynamics; LOGO: contact now → task goal later.
G1 MuJoCo Carry: 94.3% → 89.3% after robot disturbance.
🌐 https://t.co/kCQ8xynPWz

[Ziqi Han @HanZiqi666]

4/5 Recovery was tested beyond clean rollouts. In G1 MuJoCo Carry Box (60 s max; 3 × 100 trials), success was 94.3% nominal, 91.0% after a box perturbation, and 89.3% after a robot perturbation. These are simulation results.

[Ziqi Han @HanZiqi666]

3/5 LOGO separates two questions: what action is feasible now, and where should the box end up? In the Carry Box ablation, success was 51.0% without LOGO versus 94.3% with it under the same simulation setting.

[Ziqi Han @HanZiqi666]

2/5 FB (forward-backward) learning puts robot motion, object motion, and contact in a shared latent space. At inference, a task reward selects a latent command. The policy then acts in closed loop without task-specific policy retraining.

[Ziqi Han @HanZiqi666]

1/5 Why is recovery hard? During carrying, the robot, box, and contact state can all change at once. The controller may need to regain balance, re-approach the box, or keep moving toward the goal. I-BFM responds to current feedback.

[Yitang Li @li_yitang]

BTW, the real world performance is quite stable, much better than I expected from a diffusion-based controller. Huge credit to @MasterYip3 who was really the key to make this happen 🌺

[Yitang Li @li_yitang]

Excited to share our work PredActor, a predictive action diffusion policy for steerable humanoid control. This direct action-generation view is appealing because it avoids the generator-tracker mismatch. 🤩

[Yitang Li @li_yitang]

🎉BFM-Zero is accepted to ICLR 2026!

🌟In our framework, data guides exploration rather than constraining it. The robot is allowed to fall, roll, and freely explore anywhere it can reach. In unsupervised RL, no motion is inherently GOOD or BAD, and no motion should be directly encouraged or penalized by a reward function.

🚀 And inference + deploy code is just released !

1⃣ We built an ultra-lightweight FB inference tutorial and a toy MuJoCo env to show policy I/O and use cases.
2⃣ Sim2sim/deploy code is open — try out the cute and magical BFM-Zero! Full training/eval coming soon.

[Ziqi Han @HanZiqi666]

Really cool work on BFM-Zero! 🤖✨

Unsupervised RL meets humanoid control — one promptable policy for zero-shot motion tracking, goal reaching, and reward optimization.

A promising step toward general-purpose humanoid control. 🌟

[Yitang Li @li_yitang]

Meet BFM-Zero: A Promptable Humanoid Behavioral Foundation Model w/ Unsupervised RL👉 https://t.co/3VdyRWgOqb

🧩ONE latent space for ALL tasks
⚡Zero-shot goal reaching, tracking, and reward optimization (any reward at test time), from ONE policy
🤖Natural recovery & transition

[Ziqi Han @HanZiqi666]

Super cool work by my bro Ye Lei! 🔥

PredActor brings predictive action diffusion to real-world humanoid control — end-to-end, proprioception-only, and running fully onboard a Unitree G1 at 50 Hz on a Jetson Orin NX!

Really impressive work! 🚀

[Lei Ye (Raymon Yip) @MasterYip3]

Can humanoid motion generation and control be unified in a single policy?

PredActor uses a single end-to-end diffusion policy rather than separate motion generation and tracking. With proprioceptive feedback, it predicts internal future states and directly outputs actions at 50 Hz onboard a Unitree G1. Text-conditioned motion, joystick steering, disturbance response, and semantic interpolation are supported in the single policy.
Project Page: https://t.co/Z3epHOPo3x