Twitxy Saved

Perry Dong

Verified account This account is verified. Learn more

@perryadong

CS PhD @StanfordAILab, Part Time @GoogleDeepMind | Reinforcement Learning
pd-perry.github.io Joined August 2015
158 Following 1.8K Followers 82 Posts

Latest posts 5

Perry Dong Verified account @perryadong
World models have emerged as one of the biggest directions in physical AI. At the same time, RL fine-tuning is unlocking capabilities in frontier models beyond what pretraining can achieve on its own

Can we get the best of both worlds?

We propose Q-Learning with World Models (QWM)

(1/7)
8 replies 55 reposts 516 likes 32K views View on X
Perry Dong Verified account @perryadong
Pretraining has worked remarkably well across domains

We show this doesn’t hold for Q-functions in online RL from a pretrained policy — and propose IPE, a more effective way to learn Q-functions for online RL fine-tuning

(1/6)
5 replies 29 reposts 307 likes 86K views View on X
Perry Dong Verified account @perryadong
We've released the code for EXPO-FT.

Sample-efficient, reliable RL fine-tuning for VLAs — try it out today!

Code: github.com/pd-perry/expo-…
Paper: arxiv.org/abs/2605.25477
github.com GitHub - pd-perry/expo-ft Contribute to pd-perry/expo-ft development by creating an account on GitHub.
Perry Dong Verified account @perryadong May 27
Introducing EXPO-FT – Efficient, Reliable & Open-Source VLA Finetuning!

EXPO-FT unlocks π0.5 for challenging manipulation tasks:

Routing string lights & inserting the power connector to illuminate them
Striking pool ball into pocket
Inserting flower into wine bottle

(1/5)
View quoted post on X