Albert Li

39 posts

Albert Li

@albert_h_li

Robotics, Control Theory, and Machine Learning PhD Student @Caltech @AMBER_lab. @UCBerkeley '19, @Stanford '21.

Katılım Nisan 2022

33 Takip Edilen263 Takipçiler

Sabitlenmiş Tweet

Albert Li@albert_h_li·27 Haz

Super excited about 🥋judo🥋, our new open-source sampling-based MPC toolbox! 🖥️Try in under 30s: pip install judo-rai judo <click link> 🧵thread below👇

Preston Culbertson@pdculbert

🥋 We're excited to share judo: a hackable toolbox for sampling-based MPC (SMPC), data collection, and more, designed to make it easier to experiment with high-performance control. Try it: pip install judo-rai

English

3.7K

Albert Li retweetledi

Yisong Yue@yisongyue·25 Kas

My student @SaberaTalukder and I are creating a new startup that deeply rethinks how we architect and engage with multimodal models. 🚀 We are chatting with investors at #NeurIPS2025, and if you want to get on our radar, DM Sabera.

English

207

75.2K

Albert Li@albert_h_li·25 Kas

Talk to @SaberaTalukder if you want to hear refreshing ideas, spicy takes, and a broad vision for the future! Really enjoy working with her and @yisongyue at Caltech!

Yisong Yue@yisongyue

English

3.8K

Albert Li@albert_h_li·28 Haz

@Stone_Tao Thanks for the feedback! I actually have a mjwarp backend PR ready to go, but mjwarp isn’t stable yet, so it’s actually failing correctness tests for our systems vs. mujoco-C. Once this is fixed, I should very quickly be able to activate GPU support!

English

110

Stone Tao@Stone_Tao·28 Haz

@albert_h_li GPU parallelization support would be awesome. Love to try it out in our systems and in warp based ones

English

264

Albert Li@albert_h_li·27 Haz

Super excited about 🥋judo🥋, our new open-source sampling-based MPC toolbox! 🖥️Try in under 30s: pip install judo-rai judo <click link> 🧵thread below👇

Preston Culbertson@pdculbert

English

3.7K

Albert Li@albert_h_li·27 Haz

Overall, really happy with what judo has become, and thanks so much to @pdculbert and @simonlc_ for their hard work bringing this up at RAI over the past year. Additional thanks to Brandon Hung, Jiuguang Wang, and my advisor, Aaron Ames (among others!) for their support!

English

114

Albert Li@albert_h_li·27 Haz

From discussions at #RSS2025, lots of upcoming features: ⚡️ GPU support w/mjwarp 🔁 Customizable/swappable rollout/sim backends 🧠 More tasks/optimizers in the base app 🤖 Examples of hooking judo up to hardware 📦 Data collection tooling If you have ideas/feedback, let us know!

English

112

Albert Li@albert_h_li·21 Kas

Really cool work by @ziqi__ma on part-level semantic reasoning in 3D! Super useful result for dexterous manipulation.

Ziqi Ma@ziqi__ma

How far are we from embodied intelligence? We first ask whether we can build a model to locate semantic concepts in 3D. Meet Find3D🔍: a model to locate any part of any 3D object based on any text query! 👉ziqi-ma.github.io/find3dsite/ & thread below👇🧵1/4

English

354

Albert Li@albert_h_li·11 Kas

Had a great time at #Corl2024! Engaging chats over food and drink, the end of a nice collab with my friends at Stanford (shoutout @tylerlum23), and a best workshop paper award for our recent work, DROP! P.S. The DROP hardware setup has been released! 🖥️caltech-amber.github.io/drop

English

2.2K

Albert Li@albert_h_li·10 Kas

@pdculbert Reminder: you’re on this paper too!

English

124

Preston Culbertson@pdculbert·10 Kas

Congrats to @albert_h_li for winning Outstanding Paper today at the CoRL workshop for learning fine + dexterous manipulation!

English

5.1K

Albert Li@albert_h_li·8 Kas

Check out this super cool work from Tyler’s time at Nvidia - really robust, dynamic grasping with feedback!

Tyler Lum@tylerlum23

🚀Excited to present DextrAH-G at #corl2024 today! • Dynamic, reactive & robust dexterous grasping🤖🤲 • Generalizes to novel real-world objects🌎🆕 • Recovers from mistakes & retries automatically🔄💪 • Trained entirely in simulation with zero-shot sim2real transfer🎮➡️🌐

English

308

Albert Li@albert_h_li·4 Kas

I had a great time working🤝 with @tylerlum23 , @pdculbert , @krishpopdesu, Aaron Ames, @MacSchwager, and @leto__jean! 💻Website: tinyurl.com/getagrip-corl 📰Paper: arxiv.org/abs/2410.23701 🧑‍💻Code+Dataset: github.com/tylerlum/get_a… 📷Visualize the Dataset: get-a-grip.streamlit.app

English

329

Albert Li@albert_h_li·4 Kas

Our dataset size ablation found that the scale of the dataset is crucial for performance. Existing multi-finger grasp datasets with success labels only typically used ~100K grasps, far below the millions we used. And, we can keep going bigger - no diminishing returns yet!

English

343

Albert Li@albert_h_li·4 Kas

There have been many recent big grasping datasets, but few demos of real-world grasping using generative models. How do we achieve this? Introducing: Get a Grip (#corl2024)! We show that instead of generative models, discriminative models can attain sim2real transfer! 👀🧵👇

English

17.9K

Albert Li@albert_h_li·14 Eki

P.S. if you're applying for PhDs in robotics🤖, consider Cornell CS, where Preston (@pdculbert) is looking for students for his new lab! Super smart🧠, well-read on everything (probably including your favorite robotics sub-topics)📕, and will be an amazing, kind mentor🧑‍🏫.

English

131

Albert Li@albert_h_li·14 Eki

These two works done in collaboration with the incredible Preston Culbertson (@pdculbert) and Profs. Joel Burdick & Aaron Ames! 🔹FRoGGeR: arxiv.org/abs/2302.13687 🔹Intrinsic Robustness: arxiv.org/abs/2403.07249 Swing by my poster session on Wed morning to chat in person!

English

128

Albert Li@albert_h_li·14 Eki

How do we plan robust dexterous grasps quickly? At IROS '23, I proposed FRoGGeR🐸, a planner that can compute robust precision grasps in <1s. At #IROS2024, I prove why FRoGGeR works (and way more) with a new theory of ✨intrinsic robustness✨! See what it's all about 👇

English

174

Keşfet

@SaberaTalukder @yisongyue @Stone_Tao @pdculbert @simonlc_ @ziqi__ma @tylerlum23 @krishpopdesu