Log inSign up
Log inSign up
Seonghyeon Ye
295 posts
@SeonghyeonYe

Seonghyeon Ye

@SeonghyeonYe
PhD student KAIST AI (@kaist_ai) | Prev: @MSFTResearch
seonghyeonye.github.io
Joined February 2021
590
Following
1,845
Followers
RepliesRepliesRepostsRepostsMediaMedia

Log in or sign up for X

See what’s happening and join the conversation

Continue with phone
or
Log in with username or email
Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.
  • @SeonghyeonYe
    Seonghyeon Ye
    @SeonghyeonYe
    Aug 11
    In February, our release of DreamZero was among the early explorations into WAM. 6 months later, it's fantastic to see WAM truly scaling with egocentric human data from @DynaRobotics. Huge congrats to the team!
    @DynaRobotics
    Dyna Robotics
    @DynaRobotics
    Aug 10
    Today we are introducing Dyna-2, a world-action model pre-trained on one million hours of human video. At this scale, for the first time, we discovered several new scaling laws: • world-action models exhibit scaling law on human data across four orders of magnitude, from 1000
    00:00
    1
  • @SeonghyeonYe
    Seonghyeon Ye
    @SeonghyeonYe
    Aug 10
    We are organizing Scaling H2R 🦾workshop at @corl_conf this year! Submit papers & join us if you are interested in exploring human data for robot learning scaling-h2r-corl.github.io
  • @SeonghyeonYe
    Seonghyeon Ye
    @SeonghyeonYe
    Jul 2
    Check out our recent work on world models for robotics! Also presenting at ICML F2S workshop next week
    @bkjeon1211
    byeongguk jeon
    @bkjeon1211
    Jul 2
    Excited to share RoboWorld 🤖 We roll out generalist robot policies from 4,186 real initial scenes, entirely inside a video world model with no robots, and the rankings hit Pearson r = 0.989 (Spearman ρ = 0.970) with the real RoboArena leaderboard. 🧵 [1/7]
    00:00
  • @SeonghyeonYe
    Seonghyeon Ye
    @SeonghyeonYe
    Jun 23
    Training Flow Policies with RL is either unstable or not expressive enough. In our ICML paper, we introduce ⭐️Q-Flow⭐️ to tackle this challenge! I will be in-person in ICML 2026 🇰🇷. Happy to chat about robotics, world models, and beyond!
    @_jaehyeokdoo2
    JaeHyeok Doo
    @_jaehyeokdoo2
    Jun 23
    Excited to introduce our work, Q-Flow: Stable and Expressive Reinforcement Learning with Flow-Based Policy, which has been accepted to ICML 2026! By leveraging flow-consistent values, we resolve the critical trade-off between expressivity and stability in Flow-based
    00:00
    6
  • @SeonghyeonYe
    Seonghyeon Ye
    @SeonghyeonYe
    Mar 18
    Great post! Actually, it would be quite straightforward to unify WAMs and AC-WMs, similar to the UWM work by @chuning_zhu !
    @Majumdar_Ani
    Anirudha Majumdar
    @Majumdar_Ani
    Mar 17
    Article cover image
    Article
    Should world models for robotics be action-conditioned?
    A tale of two world models World models are taking off in robotics. In the past few months, we have seen the first signs of world model-based approaches surpassing vision-language-action models (VLAs)...
    1