Shuang Peng (@PSrobotics) header
Shuang Peng (@PSrobotics) profile picture

https://teapawt.net/t.co/djAEmHMExu

Joined March 2018
Photo by Shuang Peng (@PSrobotics)Photo by Shuang Peng (@PSrobotics)Photo by Shuang Peng (@PSrobotics)Photo by Shuang Peng (@PSrobotics)Photo by Shuang Peng (@PSrobotics)Photo by Shuang Peng (@PSrobotics)Photo by Shuang Peng (@PSrobotics)Photo by Shuang Peng (@PSrobotics)Photo by Shuang Peng (@PSrobotics)Photo by Shuang Peng (@PSrobotics)
Shuang Peng retweeted
Everyone trains on RoboDojo. Everyone evals on RoboDojo. We dug into what it actually simulates, and it's broken almost everywhere: material settings that never reach the physics engine, joints allowed to spin 1.5 turns each way when the real arm stops within half a turn, a task you can ace without touching anything. If your simulator is wrong, your sim2real gap is self-inflicted. Good news: read our blog post, and fix it. p-b-p-b.github.io/blog/2026/… teapawt.net/Builtin_Pb/status/2122…
What does RoboDojo really simulate? We followed it from the real world to its score, one link at a time. At almost every step, it simulates something other than what it says. Same bottle, same grasp. As shipped, it tips 78° in the hand. After our fix, it stays level.
13
16
7
180
24,524
Shuang Peng retweeted
Wanted to get into manipulation. Decided to first build a full stack that scales. Plan: copy @GeneralistAI's gripper. Fun weekend project. It was not. So @t_k_233 and I documented every design choice and lesson along the way. Meet Specialist: fully open-source hardware, data pipeline, training, and deployment code. specialist-ai.github.io/
47
113
37
906
103,340
Shuang Peng retweeted
orcs v0.1.1 is out 🚀 github.com/lok-i/orcs 🧗 perceptive whole-body locomotion task on EgoHTR: one adapted policy for rocky climbs and window crossing 🛠️ agentic UMR retargeting skill: leave frame-wise notes; your agent revises the reference
4
59
7
512
29,448
Shuang Peng 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
317
124
1,919
291,180
Shuang Peng retweeted
🤖 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 🧵👇
5
22
1
108
6,973
Shuang Peng retweeted
wanna make your motion-tracking policies perceptive? ViBe: Visual Behavior Adaptation for Perceptive Humanoid Whole-Body Control 🌐 lok-i.github.io/vibe-control… ▶️ youtu.be/aHO-7R-ZH4o 📄 lok-i.github.io/vibe-control…
5
18
2
159
29,678
Sim2realed nicely
Code is finally cleaned up and open-sourced — undergrad thesis kept me busy.😂 4 example tasks: walk/jog/run, steering, location, getup. Repo: github.com/SUZ-tsinghua/smp Try it out and give it a star! Reproduction of github.com/xbpeng/MimicKit
3
16
2
182
33,728
More sensor & debug w/ @rerundotio
2
4
53
5,962
Bothering g1 w/ @lok_s_h
1
4
1
43
3,806
Shuang Peng retweeted
Some exciting Friday news 🙂 We just open-sourced our system identification toolbox in MuJoCo 3.5. Get started today: "pip install mujoco[sysid]" mjlab v1.1 is also out featuring a brand new RGB-D renderer and now fully available on PyPI. Install with: "pip install mjlab"
12
15
3
155
10,232
I'm super excited to announce mjlab today! mjlab = Isaac Lab's APIs + best-in-class MuJoCo physics + massively parallel GPU acceleration Built directly on MuJoCo Warp with the abstractions you love.
32
139
18
863
92,671
Shuang Peng retweeted
July has been a big month for Viser! - Released v1.0.0😊 - We did some writing Some demos👇
14
102
23
758
133,404
Worked a little bit on low-level motor control. Top-down are pos, velocity, and estimate torque from current.
4
6
155
7,599
End up split into 2 canables
4
1
30
2,324