Leo Dong

@leooeld

Scaling robot intelligence @rhoda_ai Prev @Stanford

Palo Alto, CA

Joined March 2026

Leo Dong
@leooeld
Jun 20
We got our most senior “summer intern” ever😂. Welcome!
Vincent Sitzmann
@vincesitzmann
Jun 19
I am thrilled to join rhoda.ai @RhodaAI as an advisor, where I am helping harness the abilities of large-scale pre-training and video models for robotics, putting many of my lab's research learnings of the past few years into practice! I will be in-person at the
00:00
Leo Dong
@leooeld
Jun 5
yes
Jia-Bin Huang
@jbhuang0604
Jun 4
POV: attending CVPR
Leo Dong
@leooeld
May 13
I bet they used BF16-throughput as the denominator when training in FP8 or something. By that algebra, I can get you 150% MFU in no time😅. For reference, as far as I know the SOTA Hopper GEMM kernel is ~84% utilization. arxiv.org/abs/2605.05331
Leo Dong
@leooeld
Apr 30
A common misconception is that video models are too slow to run as closed-loop policies. We’ve shown that not only can they be fast, but they’re fast enough to run on a single RTX 5090! It turns out that if you co-design your model architecture and inference optimizations
Rhoda AI
@RhodaAI
Apr 30
Can a large foundation video model run as a real-time robot policy at the edge, on a single RTX 5090? • ✅ No quantization • ✅ No distillation • ✅ Full denoising (all the way from noise to clean video) We just proved it's possible. 👇🎬
00:00
Leo Dong
@leooeld
Apr 23
In-context learning is such an elegant application of a generative video model as robot policy.
Rhoda AI
@RhodaAI
Apr 23
Teaching a robot a new task typically means stopping operations, collecting teleoperated demonstrations, and retraining. That process takes hours at a minimum. We wanted to know if we could collapse it to seconds — from a single human demo, on the fly, no retraining required.
00:00