Markus Wulfmeier

Highly interesting read. Reminds me of the emergence of bang bang control in RL (Bernoulli is all you need 😉). Bet RL posttraining would be interesting... arxiv.org/abs/2610.20429

When Do Discrete Actions Make Good Continuous Controllers? arxiv.org

We need simpler RL! Great to see momentum for critic-only RL in continuous control! Here better focused expansion of our previous growing Q networks @timseyde arxiv.org/pdf/2610.20463

arxiv.org

Meanwhile our system isn't a pilot: it's in production, verifying 150k orders a day at <1% false alerts. Just like ChatGPT needed deployment feedback to become truly useful, so does physical AI 👇 nomagic.ai/the-last-few-p… Great work by Michał Grzejdziak-Zdziarski and team @no…

nomagic.ai

Thank you to every team that applied for @SPRIND's Next Frontier AI, not just the ones moving forward. Many familiar faces and two days of pitches did more to revive my hope for European AI than any strategy paper. The ambition is clear. I truly enjoyed this! 👏 x.com/i/status/…

SPRIND, Federal Agency for Breakthrough Innovation (@SPRIND) on X x.com

Excited to share that I am joining @Mila_Quebec as affiliate faculty. I will be co-advising students and collaborating on topics including robot learning, physical AI, reinforcement learning and frontier model training. 👇

Please don't use AI to answer questions when applying with us. It actually decreases chances of your application moving forward!

Spending time at #NVIDIA GTC Berlin this week! Reach out if you want to talk about mastery-first physical AI, robotics, RL, and frontier model training. More 👇

In robotics/AI and need a change of perspective? We're hiring at @ Nomagic !