Ge Yang
@EpisodeYang
Followers
3K
Following
4K
Media
218
Statuses
2K
I am planting acorns one at a time with policy gradient.
London
Joined November 2011
What can unsupervised representation learning learn from deep RL? As it turned out, we can learn representations (unsupervised) by making plans!. Check out our new work at
4
77
357
It is so cool to see in action.
Many of today's data collection systems do not capture human perceptual behaviors. The observation mismatch—between what the human sees and what the robot learns from—hinders the learning of effective manipulation policies. To see what the robot sees, we developed a VR
2
0
1
Check out this awesome demo!.
We have been focusing on policy learning for robotics for a while. But can hardware be learned as well? Check out @yswhynot ‘s recent co-design work that learns what a soft gripper should be if we want to do better manipulation.
0
0
2
RT @xuxin_cheng: Meet 𝐀𝐌𝐎 — our universal whole‑body controller that unleashes the 𝐟𝐮𝐥𝐥 kinematic workspace of humanoid robots to the phys….
0
114
0
Take a look at these amazing humanoid results. Our dreams are coming true!.
A behind-the-scenes video on how teleoperation is done. Whole-body manipulation with only a VisionPro.
0
0
1
Congratulations! Incredible team and incredible traction.
Excited to launch @DynaRobotics with a team of incredible researchers, engineers and company builders!. At Dyna, our mission is to bring affordable general-purpose AI robots to real production environments.
0
0
5
Look at how fast the training is 🔥.
Feature Splatting can now turn Objaverse assets into GS. With optimized kernel, 30K iters can be done in <1min on a single 4090 GPU.
0
0
4
So many friends in this video! Gemini robotics looks incredible. Congratulations to the team 🥳.
Meet Gemini Robotics: our latest AI models designed for a new generation of helpful robots. 🤖. Based on Gemini 2.0, they bring capabilities such as better reasoning, interactivity, dexterity and generalization into the physical world. 🧵
0
0
4
The crazy part is that their closed loop evaluation setup models potholes.
5. Front-lift and nose-dive suppression test.During hard acceleration or sudden braking, the car’s nose remains steady (no “nose lift” or “nodding”), maintaining better contact with the road.
0
0
1
lol, Jon's visualization does not disappoint : ).
I just pushed a new paper to arXiv. I realized that a lot of my previous work on robust losses and nerf-y things was dancing around something simpler: a slight tweak to the classic Box-Cox power transform that makes it much more useful and stable. It's this f(x, λ) here:
0
0
7
What really excites me about this is that Atlas vector search will become even better, making it easier for a lot of smaller teams.
We joined @MongoDB! @VoyageAI’s best-in-class embedding models and rerankers will be part of MongoDB’s best-in-class database, powering mission-critical AI applications with high-quality semantic retrieval capability. A huge thank you to everyone with us on this journey, and to
0
0
0
RT @Kimi_Moonshot: 🚀 Introducing our new tech report: Muon is Scalable for LLM Training. We found that Muon optimizer can be scaled up usin….
0
308
0
My friend did it!.
Honored to receive the Sloan Research Fellowship. Thank you for the support of the research in our lab!.
0
0
2
RT @_sholtodouglas: A distillation of our mental models that we use to think about the systems perspective on training and inference at sca….
0
34
0
This is really cool.
Last year, I joined DeepSeek with no RL experience. While conducting Mathshepherd and DeepSeekMath research, I independently derived this unified formula to understand various training methods. It felt like an "aha moment", though I later realized it was PG.
0
0
5
Pretty nice paper from Dibya! Can read side by side the R1 report :P.
4. We had signs of life back then too. It’s too easy to forget older papers. I never published my findings but a few from my friends (a biased sample):. from @avisingh599 JD, @agarwl_ . from @d_yuqing + collaborators @ Meta.
0
0
2
Check this out — they use a sparse MoE with k=2, to allow post-training pruning that reduce inference cost. It is quite clever 👏 @moritz_reuss and @jyo_pari !.
How can we make diffusion policies more computationally efficient while scaling up towards generalist policies?. Introducing MoDE: A novel generalist MoE-based Diffusion Policy
0
0
2
I really like this paper.
🤖 Introducing Human-Object Interaction from Human-Level Instructions! First complete system that generates physically plausible, long-horizon human-object interactions with finger motions in contextual environments, driven by human-level instructions. 🔍 Our approach:.- LLMs
0
0
2
Look at what @yuewang314 got to work with his students! And the best part is you can also train your robot on these data : ) and there is no VXF involved in this video :-P.
🎬Can internet videos enhance the scalability of humanoid learning?. 🤖Introducing Humanoid-X, a comprehensive dataset comprising over 20 million humanoid robot poses paired with text-based motion descriptions, on which we develop Universal Humanoid-1 (UH-1), a large model for
1
1
6