Markus Wulfmeier
@m_wulfmeier
Followers
12K
Following
10K
Media
289
Statuses
2K
Robot Intelligence - Research @GoogleDeepMind European @ELLISforEurope - priors: @oxfordrobots @berkeley_ai @ETH @MIT
London, England
Joined December 2015
Imitation is the foundation of #LLM training. And it is a #ReinforcementLearning problem! Compared to supervised learning, RL -here inverse RL- better exploits sequential structure, online data and further extracts rewards. Beyond thrilled for our @GoogleDeepMind paper! A
10
68
373
We just released the beta version of the open-source software for Reachy Mini! It means that anyone, thanks to the amazing @GoogleDeepMind mujoco simulation platform, can start building @huggingface spaces, datasets and models, even if you haven't received your robot yet.
15
65
493
Extremely useful! Transitioning to learnt NERF-based rendering was a key factor for good robot soccer sim2real transfer. https://t.co/ZNM7tWwVQD These days we would use Gaussian splatting. Thrilled for simulation's impact on whole body control in the coming years!
arxiv.org
We apply multi-agent deep reinforcement learning (RL) to train end-to-end robot soccer policies with fully onboard computation and sensing via egocentric RGB vision. This setting reflects many...
very excited about the upgrade GaussGym provides for environments for training locomotion capabilities
0
0
3
Truly enjoyed discussing the consolidation of specialist and generalist approaches to physical AI at #IROS2025 Thank you for the invitation @liu_puze @Jan_R_Peters and the who organization committee. Hoping to visit Hangzhou in physical rather than digital form myself in the
How do robots thrive in dynamic worlds? Join us at IROS LeapRiDE 2025 Full-day Workshop Oct 20, ROOM 102A, Hangzhou HIEC, China 🎯 From quadruped to humanoid ping-pong & robot soccer 🎓 Speakers from MIT, Berkeley, DeepMind, NTU, NUS, Monash, Tongji, & KU Leuven #IROS25
1
0
4
The explosion of AI capability and complexity (LLMs, VLMs, VLAs) demands better understanding. I firmly believe in studying large model behavior - from the perspective of artificial sequential decision making (like Inverse RL) to now linking it with human decision-making and
Excited to present our new paper as a spotlight talk 🌟 at the Pragmatic Reasoning in LMs workshop at #COLM2025 this Friday! 🍁 Come by room 520B @ 11:30am tomorrow to learn more about how LLMs' pluralistic values evolve over reasoning budgets and alignment 🧵
0
0
3
How do robots thrive in dynamic worlds? Join us at IROS LeapRiDE 2025 Full-day Workshop Oct 20, ROOM 102A, Hangzhou HIEC, China 🎯 From quadruped to humanoid ping-pong & robot soccer 🎓 Speakers from MIT, Berkeley, DeepMind, NTU, NUS, Monash, Tongji, & KU Leuven #IROS25
0
5
15
Check out what our amazing colleagues @GoogleDeepMind are up to in collaboration with @CFS_energy !
We’re announcing a research collaboration with @CFS_energy, one of the world’s leading nuclear fusion companies. Together, we’re helping speed up the development of clean, safe, limitless fusion power with AI. ⚛️
0
0
6
Really liked this paper about quadrotor volleyball when it came out, but I'm still waiting for the real robot videos of the full game! https://t.co/sUvT5HisiQ
sites.google.com
In this paper, we tackle the problem of learning to play 3v3 multi-drone volleyball, a new embodied competitive task requiring high-level strategic coordination and low-level agile control. The task...
0
0
1
But there is one thing I'm confused about @nathanbenaich - is there stagnation? 😉 ctrl + f 'robot' 2021 = 9 2022 = 12 2023 = 21 2024 = 37 2025 = 37
1
0
1
Truly enjoyed @nathanbenaich's State of AI 2025 report! https://t.co/im5WHniM4u Recommending the parts on robots and the applicability of ideas from robotics to LLMs/agents. (e.g. connection between sim2real to env2prod) There is much more to be learnt from robotics: on
1
1
6
You have to watch this! For years now, I've been looking for signs of nontrivial zero-shot transfer across seen embodiments. When I saw the Alohas unhang tools from a wall used only on our Frankas I knew we had it! Gemini Robotics 1.5 is the first VLA to achieve such transfer!!
22
51
337
'Papa, was ist das?' As a parent, Gemini has made my life massively easier. We found this caterpillar earlier and the answer is in fact correct (verification is much easier than generation here thanks to classical Google search). I'm very curious about how LLMs are continuing
0
0
5
Fantastic to see more examples of scalability in hierarchical RL! Simplicity and low computational cost are key to scaling these systems. I deeply believe there is more to be found, all the way back to our graphical model perspective to options
arxiv.org
We introduce Hindsight Off-policy Options (HO2), a data-efficient option learning algorithm. Given any trajectory, HO2 infers likely option choices and backpropagates through the dynamic...
Introducing Scalable Option Learning (SOL☀️), a blazingly fast hierarchical RL algorithm that makes progress on long-horizon tasks and demonstrates positive scaling trends on the largely unsolved NetHack benchmark, when trained for 30 billion samples. Details, paper and code in >
1
2
12
Great news for the European AI ecosystem! German AI start-up in funding talks at $4bn valuation -
ft.com
Black Forest Labs explores raising between $200mn and $300mn
1
0
2
The position paper track at #NeurIPS2025 was a great idea, the acceptance rate of under 6% not so much! This is unnecessarily low and will reduce interest in any future iterations of the track. (Disclaimer: our team is part of the 94%) @NeurIPSConf
0
0
17
If you want one brain for every robot, cross-embodiment learning is the key. Check out the new model and tech report for sparks of it https://t.co/Zb0gCRSdg5 And catch up with the team (unfortunately without me this year) at #CORL2025!
New Gemini Robotics 1.5 models will enable robots to better reason, plan ahead, use digital tools like Search, and transfer learning from one kind of robot to another. Our next big step towards general-purpose robots that are truly helpful — you can see how the robot reasons as
0
2
7
Finally! Super nice results, stacking and building with Lego has never been an practical application but a sign of dexterity and precise control - a sign of things becoming possible.
At Generalist, we’re working towards a future where robots can do anything. To that end, the robots build now, too. We’ve trained a robot to do one-shot assembly, constructing Legos end-to-end: no custom engineering, just pixels in → Lego copies out.
1
0
10
It's great to see our work pushing the frontier of scientific discovery - to boldly go where no one has gone before!
Using AI to advance our understanding of fundamental physics is the dream. Excited to see our latest AI model 'Deep Loop Shaping' help @LIGO and @Caltech detect the gravitational waves of intermediate-mass black holes better! Published in @ScienceMagazine
1
0
7