Kun Lei (@kunlei15) / X

Kun Lei

266 posts

Kun Lei

@kunlei15

Research on reinforcement learning and robotics. RL-100, Uni-O4, BPPO

Joined October 2022

Pinned
Kun Lei
@kunlei15
Nov 9, 2025
RL feels messy, but a two-axis view—data source (on/off/offline) × update schedule (one-step/multi-step/iterative)—brings order. I wrote a post unifying them with shared equations. lei-kun.github.io/blogs/rl.html Robotic FMs (e.g., GEN-0, pi_0.5) grow via a data flywheel. Best fit:
Kun Lei
@kunlei15
Oct 16, 2025
Introducing RL-100: Performant Robotic Manipulation with Real-World Reinforcement Learning. lei-kun.github.io/RL-100/ 7 real robot tasks, 900/900 successes. Up to 250 consecutive trials in one task, running 2 hours nonstop without failure. High success rate against physical
00:00
Kun Lei
@kunlei15
Oct 23, 2025
Real-world deployment is crucial for robotics. I wrote about a recipe: strong base model + RL post-training (offline & online) → toward human-level capability. lei-kun.github.io/blogs/RL100.ht…
Kun Lei
@kunlei15
Oct 12, 2025
Robots level up tomorrow — come see real-world reinforcement learning teach them new tricks.
Kun Lei
@kunlei15
Nov 7, 2023
On-policy RL is all you need! Uni-O4: Unifying Online and Offline Deep Reinforcement Learning with Multi-Step On-Policy Optimization. #CoRL2023 Deployable Workshop lei-kun.github.io/uni-o4/ (1/n
00:00
Kun Lei
@kunlei15
Oct 17, 2025
From demo to duty: RL-100 can serve continuously for ~7 hours—reliable, real-world robot helps.
00:00
03:26
Kun Lei
@kunlei15
Oct 16, 2025
Introducing RL-100: Performant Robotic Manipulation with Real-World Reinforcement Learning. lei-kun.github.io/RL-100/ 7 real robot tasks, 900/900 successes. Up to 250 consecutive trials in one task, running 2 hours nonstop without failure. High success rate against physical
Kun Lei
@kunlei15
Oct 20, 2025
We are releasing per-task point-cloud trajectories. Visualizations highlight why tight crop boxes matter for 3D reps: attend to object–robot contact and drop irrelevant details. Cleaner signals → stronger policy learning.
00:00
03:26
Kun Lei
@kunlei15
Oct 16, 2025
Introducing RL-100: Performant Robotic Manipulation with Real-World Reinforcement Learning. lei-kun.github.io/RL-100/ 7 real robot tasks, 900/900 successes. Up to 250 consecutive trials in one task, running 2 hours nonstop without failure. High success rate against physical
Kun Lei
@kunlei15
Mar 7, 2024
🌈🌈🌈We have released all the code, real-world robot training setup, hardware, dataset, and checkpoints on github.com/Lei-Kun/Uni-O4 🚀🚀🚀Run and train the robot in your lab within several minutes.
Kun Lei
@kunlei15
Nov 7, 2023
On-policy RL is all you need! Uni-O4: Unifying Online and Offline Deep Reinforcement Learning with Multi-Step On-Policy Optimization. #CoRL2023 Deployable Workshop lei-kun.github.io/uni-o4/ (1/n
00:00
Kun Lei
@kunlei15
Oct 24, 2025
Add 1x speed video in the project web. Come to count if you are suffering from insomnia. lei-kun.github.io/RL-100/#100suc…
03:26
Kun Lei
@kunlei15
Oct 16, 2025
Introducing RL-100: Performant Robotic Manipulation with Real-World Reinforcement Learning. lei-kun.github.io/RL-100/ 7 real robot tasks, 900/900 successes. Up to 250 consecutive trials in one task, running 2 hours nonstop without failure. High success rate against physical
Kun Lei
@kunlei15
Jan 17, 2024
Uni-O4 got accepted by #ICLR2024 ~ Let Uni-O4 train/fine-tune your robots with minimal interactions in real-world 🚀🚀🚀 See you in Vienna🚣‍♂️🚣‍♂️🚣‍♂️
Kun Lei
@kunlei15
Nov 7, 2023
On-policy RL is all you need! Uni-O4: Unifying Online and Offline Deep Reinforcement Learning with Multi-Step On-Policy Optimization. #CoRL2023 Deployable Workshop lei-kun.github.io/uni-o4/ (1/n
00:00
Kun Lei
@kunlei15
Feb 4, 2025
Supervised learning (BC) is great, but also limited by the dataset. How about BC + RL finetuning (offline & online)? It’s the way to surpass the dataset & human performance. See our work use this pipeline: lei-kun.github.io/uni-o4/
Ignat Georgiev
@imgeorgiev
Feb 3, 2025
Behavior Cloning (BC) has been the new hot thing in #Robotics for the past year. I finally sucked my teeth into it and tried to decipher why it has worked so well for problems where RL struggles imgeorgiev.com/2025-01-31-why… Let me know if you have other interesting perspectives!
Kun Lei
@kunlei15
Sep 4, 2025
Very impressive！Focused sim2real mobile dexterous manipulation，human data is leveraged for the guidance in reinforcement learning.
Zhecheng Yuan
@fancy_yzc
Sep 3, 2025
👐How can we leverage multi-source human motion data, transform it into robot-feasible behaviors, and deploy it across diverse scenarios?  👤🤖Introduce 𝐇𝐄𝐑𝐌𝐄𝐒: a versatile human-to-robot embodied learning framework tailored for mobile bimanual dexterous manipulation.
00:00
Kun Lei
@kunlei15
Apr 1, 2024
Robot dogs can achieve locomotion and manipulation with their 🦵legs 🦵 like some animals. We train the dog to complete some daily tasks using their 🦵legs🦵during navigation🚀.
Zhengmao He
@ZhengmaoHe
Apr 1, 2024
Quadruped robots already have four manipulators (aka legs), do we really need extra arms for them? 🤔 Dive into our latest work: quadruped robots handle complex loco-manipulation tasks autonomously with just legs! 🐾 zhengmaohe.github.io/leg-manip
00:00
Kun Lei
@kunlei15
Dec 31, 2023
Happy New Year! Throughout the incredible year 2023, I had the opportunity to connect with numerous exceptional researchers. Every recognition from my peers motivates me to continue delivering more solid and accessible work.