Brent Yi
@brenthyi
Followers
1K
Following
1K
Media
10
Statuses
116
PhD student @berkeley_ai ๐
Joined November 2009
๐ฎโ๐จ๐ค๐ฅ Tired of building dexterous tasks by hand, collecting data forever, and still fighting with building the simulator environment? Meet GenDexHand โ a generative pipeline that creates dex-hand tasks, refines scenes, and learns to solve them automatically. No hand-crafted
1
24
71
Manifold Muon stabilizes large model training, but it's expensive ๐ฐ -- requiring an inner loop solve for each update. ๐ก But you can significantly accelerate it, leading to 2.3x speedup on @thinkymachines's experiment with no performance loss! Blog and ๐งตbelowโฆ
5
25
268
@brenthyi and I are at #IROS2025 to present PyRoki ๐ซถ Find us either: - tomorrow (tues 10:30am) in room 102C, or - anytime during the conference!
Excited to introduce PyRoki ("Python Robot Kinematics"): easier IK, trajectory optimization, motion retargeting... with an open-source toolkit on both CPU and GPU
1
3
67
Everyone says they want general-purpose robots. We actually mean it โ and weโll make it weird, creative, and fun along the way ๐ Recruiting PhD students to work on Computer Vision and Robotics @umdcs for Fall 2026 in the beautiful city of Washington DC!
28
73
469
We open-sourced the full pipeline! Data conversion from MimicKit, training recipe, pretrained checkpoint, and deployment instructions. Train your own spin kick with mjlab: https://t.co/KvNQn0Edzr
github.com
Train a Unitree G1 humanoid to perform a double spin kick using mjlab - mujocolab/g1_spinkick_example
7
78
389
It was a joy bringing Jasonโs signature spin-kick to life on the @UnitreeRobotics G1. We trained it in mjlab with the BeyondMimic recipe but had issues on hardware last night (the IMU gyro was saturating). One more sim-tuning pass and we nailed it today. With @qiayuanliao and
Implementing motion imitation methods involves lots of nuisances. Not many codebases get all the details right. So, we're excited to release MimicKit! https://t.co/7enUVUkc3h A framework with high quality implementations of our methods: DeepMimic, AMP, ASE, ADD, and more to come!
25
93
653
We wrote a book about representation learning! Itโs fully open source, available and readable online, and covers everything from theoretical foundations to practical algorithms. ๐ทโโ๏ธ Weโre hard at work updating the content for v2.0, and would love your feedback and contributions
13
204
1K
I've recently started working on humanoids and this project set the bar super high for me ๐ Very impressive demo, and our team is working on many interesting projects, exciting time!
Humanoid motion tracking performance is greatly determined by retargeting quality! Introducing ๐ข๐บ๐ป๐ถ๐ฅ๐ฒ๐๐ฎ๐ฟ๐ด๐ฒ๐๐ฏ, generating high-quality interaction-preserving data from human motions for learning complex humanoid skills with ๐บ๐ถ๐ป๐ถ๐บ๐ฎ๐น RL: - 5 rewards, - 4 DR
1
4
43
New project from @kevin_zakka I've been using + helping with! Ridiculously easy setup, typed codebase, headless vis => happy ๐
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.
1
3
37
Congratulations to the videomimic team for winning the best student paper award at CoRL 2025 ๐ฅน๐ Grateful to the CoRL community for the recognition!
our new system trains humanoid robots using data from cell phone videos, enabling skills such as climbing stairs and sitting on chairs in a single policy (w/ @redstone_hong @junyi42 @davidrmcall)
6
23
250
๐ฎ
Humanoid motion tracking performance is greatly determined by retargeting quality! Introducing ๐ข๐บ๐ป๐ถ๐ฅ๐ฒ๐๐ฎ๐ฟ๐ด๐ฒ๐๐ฏ, generating high-quality interaction-preserving data from human motions for learning complex humanoid skills with ๐บ๐ถ๐ป๐ถ๐บ๐ฎ๐น RL: - 5 rewards, - 4 DR
0
0
6
How can we enable finetuning of humanoid manipulation policies, directly in the real world? In our new paper, Residual Off-Policy RL for Finetuning BC Policies, we demonstrate real-world RL on a bimanual humanoid with 5-fingered hands (29 DoF) and improve pre-trained policies
8
50
230
It's been over 2 years since we first released GELLO. Weโve updated the project with 3 new GELLOs to control the YAM, ARX and similar arms! Enables easy data collection for policy training (ie finetuned Pi0!). All open source contributions from @jeffreyliu05 and @hmahesh007!
5
24
148
Introducing Masquerade ๐ญ: We edit in-the-wild videos to look like robot demos, and find that co-training policies with this data achieves much stronger performance in new environments. โNote: No real robots in these videosโItโs all ๐ช๐ผ โก๏ธ ๐ฆพ ๐งต1/6
18
38
280
๐ขThrilled to share that I'll be joining Harvard and the Kempner Institute as an Assistant Professor starting Fall 2026! I'll be recruiting students this year for the Fall 2026 admissions cycle. Hope you apply!
We are thrilled to share the appointment of @QianqianWang5 as an #KempnerInstitute Investigator! She will bring her expertise in computer vision to @Harvard. Read the announcement: https://t.co/Aoh6A5gp9B
@hseas #AI #ComputerVision
101
41
724
Want to achieve extreme performance in motion trackingโand go beyond it? Our preprint tech report is now online, with open-source code available!
36
241
1K
Excited to open-source GMR: General Motion Retargeting. Real-time human-to-humanoid retargeting on your laptop. Supports diverse motion formats & robots. Unlock whole-body humanoid teleoperation (e.g., TWIST). video with ๐
22
113
697
SPI-Active is accepted by CoRL2025, see you at Seoul ^โ_โ^ https://t.co/NmgvDg08rv
How can we align simulation with real-world dynamics for legged robots? Check out our new work: SPI-Active โ Sampling-based system identification with active exploration for sim-to-real transfer in legged systems. We leverage sampling-based optimization to estimate robot
0
4
43
"Cameras as Relative Positional Encoding" TLDR: comparison for conditioning transformers on cameras: token-level raymap, attention-level relative pose encodings, a (new) relative encoding Projective Positional Encoding -> camera frustums, (int|ext)insics for relative pos encoding
2
51
465
๐ค Initial code release is up for VideoMimic Real2Sim! https://t.co/jKGM3lyRCq VideoMimic is a real-to-sim-to-real pipeline for deploying humanoids in the real world. It supports: - Human motion capture from video - Environment reconstruction for simulation from video -
github.com
Visual Imitation Enables Contextual Humanoid Control. arXiV, 2025. - hongsukchoi/VideoMimic
our new system trains humanoid robots using data from cell phone videos, enabling skills such as climbing stairs and sitting on chairs in a single policy (w/ @redstone_hong @junyi42 @davidrmcall)
3
35
166