Antoine Cully
@CULLYAntoine
Director of the Adaptive and Intelligent Robotics Lab (AIRL) Reader (Associate Professor) in Robotics and AI at Imperial College London
My lab is Hiring! I am looking for a research engineer (pre doc) or a postdoc with strong skills in computer vision and state estimation to join our TRUSTLINE project. More info: imperial.ac.uk/jobs/search-jo…
For the fifth consecutive year, we are coming back from from @GeccoConf with a Best Paper Award!!!🥇 congratulations to my former PhD student @MFlageat and our great collaborators Johann Huber, Francois Helenon, and @doncieux 🎉🦎
Want to hear about a new approach that scales with massive parallelization, generating diverse, high-performing DNN policies in minutes? Join our presentation at @GeccoConf on Friday 18th July 11:00am! Work done with @maxencefaldor & @CULLYAntoine! #GECCO2025
🦎 Thank you all for a great first edition of the workshop "Evolving self-organisation" yesterday at @GeccoConf It's been a great mix of artificial life and GECCO vibes. Especially thanks to our organisers @zzznah @risi1979 @eyvindn @RandazzoEttore @miltonllera @JoachimWinther
"When does Neuro-evolution out-compete Reinforcement Learning in Transfer Learning tasks?" Find us at @GeccoConf where we are presenting this work of our Grow-AI team with @risi1979 @miltonllera @JoachimWinther @eplantec arxiv.org/abs/2505.22696 sebastianrisi.com/grow-ai/
Checkout this paper where we introduce a new neuroevolution approach for Quality-Diversity, which is as effective as actor-critic based methods and as scalable as the standard MAP-Elites. This means we can generate thousands of high-performing policies in minutes!
Want to hear about a new approach that scales with massive parallelization, generating diverse, high-performing DNN policies in minutes? Join our presentation at @GeccoConf on Friday 18th July 11:00am! Work done with @maxencefaldor & @CULLYAntoine! #GECCO2025
Excited to introduce this new paper that opens the door for unsupervised multi-objective Quality-Diversity algorithms! This is impactful to produce large collections of complex designs even in scenarios where the best descriptor definition is unknown!
🚨 New paper at @GeccoConf with @CULLYAntoine! 🦎 We introduce MOUR-QD — the first multi-objective QD algorithm that works without a hand-crafted feature space. A step toward more flexible, unsupervised exploration. 🧵👇
🚀 QDax v0.5.0 just released!! github.com/adaptive-intel… Huge thanks to @p_templier @maxencefaldor @hannah_janmo @MFlageat @lisa_coiffard Runjun Mao and @CULLYAntoine for making QDax better for the entire QD community! A 🧵 with some new functionalities 👇
Very proud to receive this award in honor of Julian! I was unable to go to Evostar this year to get the award physically (already an important commitment on that exact day) but I will next year!
Julian Francis Miller Award 2025 🏆 At @EvostarConf 2025, we were thrilled to announce this year’s awardee: Jean-Baptiste Mouret! 🎉 Congratulations 👏
Very cool work! Fun to see an algorithm inspired by MAP-Elites help advance numerous open problems in math, computer science, AND improve Google's production infrastructure (data centers, TPU design, and AI training)! (Paraphrasing their paper and tweet below). Great work…
After 1.5 years of work, I'm so excited to announce AlphaEvolve – our new LLM + evolution agent! Learn more in the blog post: deepmind.google/discover/blog/… White paper PDF: storage.googleapis.com/deepmind-media… (1/2)
Google's AlphaEvolve system discovers and optimises algorithms by orchestrating Large Language Models (LLMs) within an evolutionary computation framework inspired MAP Elites from @jb_mouret and @jeffclune. LLMs propose code modifications which are then automatically evaluated,…
The most important question of the next few years.
However, significant hurdles remain. As @robertarail discussed, it's unclear if current LLMs possess the intrinsic capability for sustained open-ended improvement. And evidence suggests even SOTA agents (e.g. AIDE @zhengyaojiang) plateau, unlike self-improving humans. The…
🚀 2 days remaining to apply for the Imperial Robotics Summer School! 📄 Applications close on the 2nd May 2025! Apply now at imperial.ac.uk/robotics/imper… #ImperialRoboticsSummerSchool
🚨 Upcoming seminar! Our next session features Mark Sagar, a double Academy Award winner in Scientific and Engineering. Mark is known from his participation in blockbusters such as Avatar, King Kong, and Spiderman 2. His work is set in the intersection of AI, Interactive Human…
The variety of top speakers and diverse topics in our seminars is always a blast, but having a double-Oscar winning scientist feels extra special! We need to do in-person only to show more juicy stuff from the movies 😱 I know some of our regular audience will need to travel…
🚨 Upcoming seminar! Our next session features Mark Sagar, a double Academy Award winner in Scientific and Engineering. Mark is known from his participation in blockbusters such as Avatar, King Kong, and Spiderman 2. His work is set in the intersection of AI, Interactive Human…
Same thing on this side!! Looking forward to the next opportunity to collaborate again!
💯I couldn't agree more 😊
awesome collaborators! @maxencefaldor, @CULLYAntoine, @jeffclune 🌟 project:
I am thrilled to introduce OMNI-EPIC: Open-endedness via Models of human Notions of Interestingness with Environments Programmed in Code. Led by @maxencefaldor and @jennyzhangzt, with @CULLYAntoine and myself. 🧵👇
💯I couldn't agree more 😊
awesome collaborators! @maxencefaldor, @CULLYAntoine, @jeffclune 🌟 project:
Great talk from @_rockt and @jparkerholder this morning at #ICLR2025, again super packed!! Did not have time for my question so here it goes: it seems that the multi-agent aspect is conditioned on the user inputs, but to go further we need to give the world its own mind. How?
I'm giving a talk at 11:30 today in the #ICLR World Models workshop. "Open-ended Agent Learning in the Era of Foundation Models and Foundation World Models." Drop by if you are interested! sites.google.com/view/worldmode…