Ran Cheng

Research Director @Primebot, Ex Embodied Intelligence @Lingbot, RobbyAnt. Ex-Head of AI @Midea. Ex Huawei Noah’s Ark Lab. McGill CIM alum.

[Ran Cheng @RanCheng10]

Stumbled on Reversal Q-Learning today and got a little obsessed — it's one of those ideas that's so elegant it feels obvious after you see it.

Spent the whole afternoon building an interactive explainer to do it justice. 12 little demos you can drag and play with 👇
https://t.co/6CMxL9gozS

[Ran Cheng @RanCheng10]

@jianyuan_wang that's a piece of art, TBH!

[Ran Cheng @RanCheng10]

@E0M @chris_j_paxton Impressive! Could the passive elasticity of this gripper eliminate the need for force feedback control signals, since force can in some cases be measured visually?

[Ran Cheng @RanCheng10]

Geometry is dead in robot learning, we no longer need any coordinate systems for robot to learn the world, we need aligned modalities!

[Ran Cheng @RanCheng10]

🚀 VAMPO is accepted to ICML 2026!

Robots should not just predict plausible videos — they need to predict the right visual dynamics for action.

We turn video diffusion denoising into policy optimization, improving future prediction for robot control.

Paper: https://t.co/VPcPimdwdN
Website: https://t.co/IMwfI9M1m7

[Ran Cheng @RanCheng10]

Crazy thought: Has anyone tried using Classical Chinese (文言文) to maximize LLM token efficiency? 📜🤔
It’s essentially an ancient, extreme form of data compression. In the agentic era where context windows and token costs dictate performance, could this ultra-dense language be the ultimate practical hack for agent-to-agent communication?
Would love to read any research or experiments on this! 👇
#LLMs #AIAgents #NLP #TokenEfficiency #MachineLearning #AI

[Ran Cheng @RanCheng10]

As a $20 claude user, I feel like I have hired an amazing engineer who takes long breaks frequently, and always has a 4 day weekend. 🤷‍♂️

source: jepace (reddit)

[Ran Cheng @RanCheng10]

@zhengyiluo @KC_Q1015 reminds me of the good old days of “SayPlan” and “SayCan”

[Ran Cheng @RanCheng10]

@AjdDavison guess they need some in house SLAM team for accurate pose estimation for their UMI data

[Ran Cheng @RanCheng10]

@zhengyiluo What happend to the trash can tho🤔

[Ran Cheng @RanCheng10]

https://t.co/XhNxtxUulK

[Ran Cheng @RanCheng10]

@haotian_yeee Hell yeah! please integrate to LeRobot @ClementDelangue

[Junyang Lin @JustinLin610]

https://t.co/mgVhhZVTKg

[Oier Mees @oier_mees]

If you missed @chichengcc's guest lecture on "Robotics: Beyond Algorithms" from my @ETH robot learning course, check it out on YouTube! He shares insights that are rarely taught & hard to learn in academia.
📽️ YouTube: https://t.co/coRqOHHB3n
📚 Course: https://t.co/QJcfXJRfX8

[Ran Cheng @RanCheng10]

Time is the only currency you can’t earn more of.

For humans, tokens aren’t words — they’re moments of life.

Spend them like they matter.

[Ran Cheng @RanCheng10]

@DJiafei Human preference can bottleneck robot learning. First learn the world as it is; only then optimize for what humans want.

[Ran Cheng @RanCheng10]

Robots do not care about success or failure. They care about learning the world. Success is just human preference imposed on top of dynamics.

Human preference is not the foundation of robot learning—it is a bottleneck. The robot should first learn the world, then learn what we want.

[Anirudha Majumdar @Majumdar_Ani]

https://t.co/3tzZESvF4L

[Ran Cheng @RanCheng10]

https://t.co/F2AaFvjyix

[Ran Cheng @RanCheng10]

Why do some continual learning methods forget everything while others don't?

We now have a single number that explains it: Context Channel Capacity (C_ctx).

📐 Zero forgetting ⟺ C_ctx ≥ H(T)
🔺 Impossibility Triangle: zero forgetting + online learning + finite params → pick 2
🧠 HyperNets bypass the triangle entirely by redefining params as functions, not states

Validated across 1,130+ experiments on 8 CL methods. C_ctx perfectly predicts who forgets and who doesn't.

📄 https://t.co/MyIXCQJORM

[Ran Cheng @RanCheng10]

@bercankilic learn how to forget is very hard, I've tried 1800+ experiments with my agent teams, failed and failed again, here's my failure summaries: https://t.co/nR72EYNUWo if you want to waste your time reading it.