Lizhi(Gary) Yang

Ph.D. candidate in Robotics & Control @Caltech

[Lizhi(Gary) Yang @lzyang2000]

Over the weekend coded up an implementation of TWIST2 in MjLab in order to update the framework from the EOL IsaacGym and it is fully public on GitHub: https://t.co/kzLLKH1n9T.

Credits to @ZeYanjie and @kevin_zakka for their amazing work!

[Lizhi(Gary) Yang @lzyang2000]

@FengLeader @chris_j_paxton We have a limitations section ;)

[Lizhi(Gary) Yang @lzyang2000]

@PTrubey @chris_j_paxton For me this is more to show that we don’t need full body motion generation to still accomplish tasks - a starting point to build up from, since you can delegate a lot of the higher level planning to existing vlms with “common sense”, i.e. they provide good sparse waypoints

[Lizhi(Gary) Yang @lzyang2000]

@chris_j_paxton Thanks for the shoutout :) Everything is open sourced at https://t.co/LibcKNVpgT
We look forward to making it behave a lot better!

[Lizhi(Gary) Yang @lzyang2000]

@chris_j_paxton To be fair I think people use it cuz it’s a UMI clone haha (not that it’s bad)
I feel wrist cameras are for vision coverage because current vla-style models rarely have memory

[Lizhi(Gary) Yang @lzyang2000]

Doing a little exercise @AMBER_lab

[Lizhi(Gary) Yang @lzyang2000]

@diydatascience @kyleichan They were on a minibus, pretty sure they can keep up haha

[Lizhi(Gary) Yang @lzyang2000]

@chris_j_paxton They just finished - the top autonomous team completed the half marathon in around 50 minutes, beating the human WR by 7 min

[Jianuo Cao @JianuoCao]

1/ Introducing CLAW 🦀 — a pipeline for scalable generation of language-annotated whole-body motion data for the Unitree G1 humanoid.
Joint work with @ThomasYuxinChen at MSC Lab, UCberkeley
Code: https://t.co/gN45zcSyBX
Tech Report: https://t.co/AtpuSIQjj7

[Lizhi(Gary) Yang @lzyang2000]

@chris_j_paxton @chen_hongyi_ I remember this being one off the earliest works - the idea of learning inverse kinematic/dynamics has gone far

[Lizhi(Gary) Yang @lzyang2000]

Hardware deployment pipeline is released. Our robot is worked pretty hard so might not be most best indicator of performance.

[Lizhi(Gary) Yang @lzyang2000]

Another weekend another update! Now with support for the BONES-SEED dataset from @TheBonesStudio and @NVIDIARobotics

[Yuke Zhu @yukez]

We've just open-sourced the SONIC training code, the training data, the algorithms used to generate the data, and more to come. You now have the full recipe for building SONIC whole-body controllers for your own humanoids. Enjoy!

Code: https://t.co/WAZ1P12shu
Web Demo: https://t.co/sc3yxIKpuJ

[Kevin Zakka @kevin_zakka]

mjlab v1.3.0 is out! It's packed with features: a revamped viewer built on mjviser, a preset-based terrain system, simplified actuator config, and new MDP primitives. Here's a Go1 quadruped trained in the stair terrain with the new terrain-aware foot height sensor.

[Lizhi(Gary) Yang @lzyang2000]

The repo now includes a GPU cache and full cpu offloading of the motion dataset as a last resort, hopefully no one ever needs that.

[Lizhi(Gary) Yang @lzyang2000]

Another weekend another update! Now with support for the BONES-SEED dataset from @TheBonesStudio and @NVIDIARobotics

[Lizhi(Gary) Yang @lzyang2000]

Over the weekend coded up an implementation of TWIST2 in MjLab in order to update the framework from the EOL IsaacGym and it is fully public on GitHub: https://t.co/kzLLKH1n9T.

Credits to @ZeYanjie and @kevin_zakka for their amazing work!

[Lizhi(Gary) Yang @lzyang2000]

@carlosdponx Also tried infusing safety in the learning controller itself: https://t.co/e70bjRxJEs

I try to bring all of my robots out and let them run in the real world

[Lizhi(Gary) Yang @lzyang2000]

@carlosdponx Our lab actually does a lot of safety, for example https://t.co/VZ9ieci9E2 and https://t.co/e70bjRxJEs

[Lizhi(Gary) Yang @lzyang2000]

@chris_j_paxton It used to be (and I argue still is) the lynchpin holding everything together

[Lizhi(Gary) Yang @lzyang2000]

@chris_j_paxton I think it’s gone from his website now

[Lizhi(Gary) Yang @lzyang2000]

I might just do robot repair after I graduate… #Robotics