Looking forward to attending the European Graduate Network Symposium 2024 at the Honda Research Institute in Germany next week! Honda is an important partner for our research group and we're working together on exciting new projects. One of these is a new JAX-accelerated framework for RL and multi-agent RL targeting real-world applications, and we can't wait to share this with the research community soon. Spoiler: it's fast!⚡

Looking forward to teaching at the Multi-Agent Reinforcement Learning (MARL) summer school hosted by @IIIACSIC in Barcelona from 16-19th September. Registration opens tomorrow and in-person spots are limited! Why should you join? MARL is at the cutting edge of machine learning, enabling a collective of agents to learn optimal interactions in shared environments. From autonomous driving and multi-robot factories to automated trading and energy network management, the applications of MARL are revolutionizing modern life. In this course I will be following our new MARL MIT Press textbook, freely available at marl-book.com During this course, you will: - Gain a solid introduction to the foundations of multi-agent reinforcement learning theory and algorithms. - Explore interactive game models and different solution concepts. - Dive into contemporary MARL algorithms, including centralized training with decentralized execution, value decomposition, and self-play. 📅 Key Details: - Location: Barcelona @ IIIA-CSIC, Campus UAB - Dates: 16-19th September (both included) 10am-1:15pm - Registration: Opens tomorrow August 27 - Format: In-person (limited availability) and online 🔗 Register here: iiia.csic.es/en-us/marl-cour… #MARL #ReinforcementLearning #SummerSchool #Barcelona #MachineLearning #AI #Education #MITPress

🎉 2x RLC papers and 1x RLC workshop paper! Delighted to share that we have had 3 papers accepted to the first Reinforcement Learning Conference @RL_Conference. 1. Multi-view Disentanglement for Reinforcement Learning with Multiple Cameras We propose a self-supervised auxiliary task to learn a disentangled representation from multiple cameras that is robust to a reduction in available cameras 📷 arxiv.org/abs/2404.14064 2. Planning to Go Out-of-Distribution in Offline-to-Online Reinforcement Learning We propose a planning paradigm for the offline-to-online setting that targets data for collection during the online phase that are unlikely under the behavior policy's occupancy measure arxiv.org/abs/2310.05723 3. Agent-Temporal Credit Assignment for Optimal Policy Preservation in Sparse Multi-Agent Reinforcement Learning [CoCoMARL Workshop] We propose a new reward redistribution function that aids in agent-temporal credit assignment while preserving optimal policy guarantees! 🚀💡 openreview.net/forum?id=dGS1… Congratulations to @trevormcinroe @MhairiDunion @KaliTessera and @_AdityaKapoor_ for their great work! Please get in touch with them if you would like to find out more about the work. Myself, @MhairiDunion, and @Trevormcinroe will be at RLC and are happy to meet. #RLC2024 #RL #ReinforcementLearning

🎉 We have an exciting update to our open-source MARL libraries for you! Our popular EPyMARL library (439⭐) with highly customisable implementations for many MARL algorithms has now become even better - welcome to EPyMARL v2.0! github.com/uoe-agents/epymar… See our new blog post for more details: agents.inf.ed.ac.uk/blog/epy… In short, what's new? - Training in environments with individual rewards for all agents - Logging to Weights and Biases - SMACv2, SMAClite, and PettingZoo environments - Plotting script to quickly visualise your runs - Migration from deprecated Gym 0.21 library to Gymnasium The dependency on the deprecated Gym library increasingly became a pain point in using EPyMARL and many of our environments. No more such installation headaches, compatibility wrappers, and other issues, since not just EPyMARL but also our most popular environments got a facelift to migrate them to Gymnasium, including: - LBF: github.com/uoe-agents/lb-for… - RWARE: github.com/uoe-agents/roboti… - SMAClite: github.com/uoe-agents/smacli… - Matrix games: github.com/uoe-agents/matrix… Please check it out and let us know what you think!

Excited to head to Prague next week for the @ieeeras Summer School on Multi-Robot Systems (July 29 - Aug 2)! 🌍🚀 I'll give a lecture on Multi-Agent Reinforcement Learning (MARL), based on our new @mitpress textbook marl-book.com Details: mrs.fel.cvut.cz/summer-schoo…

Postdoc application closing soon!

New PhD position in Multi-agent RL funded by @Leonardo_live. Start between Sep'24 to Jan'25, funded for 3.5 years, working with our group and Leonardo. Candidate must *not require ATAS academic-technology-approval…. To apply/questions: email CV, transcripts, and research ideas to me.

Postdoc position in multi-modal RL and LLMs at School of Informatics, University of Edinburgh. International applicants are welcome. Deadline 23 July 2024. Details here: elxw.fa.em3.oraclecloud.com/… @InfAtEd @EDINrobotics

[1/6] After fantastic visits recently to London and Edmonton, Canada, I am excited to announce my next stop is China! 🌏 From June 10-21 I will present to universities, businesses, and UK embassy in Beijing, Shanghai, Shenzhen, and Hong Kong. See thread for schedule details⬇️

Excited to be giving a main stage talk at the #UpperBound2024 AI conference by Amii @AmiiThinks next week in Edmonton, Canada! My talk will explore current research in deep RL and LLM agents. Talk details here: upperbound.ai/speakers/SPEWO… Happy to meet people, please DM/email me.

Excited to announce that China Machine Press will publish the Chinese translation of our new MARL textbook in Dec 2025 📚 Looking forward to more audiences being able to reach the fascinating world of Multi-Agent Reinforcement Learning 🎉 @mitpress marl-book.com

UPDATE: The copyedited version (formatting edits, index list) of the MARL book is now online! marl-book.com Next is the book cover design and then the book will go to print and stores in late 2024! Lecture slides will be released in coming weeks. @mitpress

Our 2018 survey in AIJ on Agent Modelling with @PeterStone_TX has reached 500 citations! 🎆 Five years after publication, many of the open problems listed in Sec5 are still wide open, esp "Autonomous Model Contemplation and Revision". Lots of PhD topics! doi.org/10.1016/j.artint.201…

For people interested in explainable/trustworthy AI, with focus on #AutonomousDriving, we have released a new survey! Explainable AI for Safe and Trustworthy Autonomous Driving: A Systematic Review arxiv.org/abs/2402.10086 @itssieee @XAI_Research

1/7 Great start in 2024! New papers from our group: TMLR @TmlrOrg, 2x AAMAS24 @AAMASconf, AAAI24 @RealAAAI, ICLR24 @iclr_conf, ICRA24 @ieee_ras_icra. PLUS new Workshop on Ad Hoc Teamwork at AAAI24 @RealAAAI sites.google.com/view/ad-hoc…. Titles and links in this thread.

Wikipedia says the model I developed in my PhD, Stochastic Bayesian Game (SBG), has been applied in a range of problems. This is very cool🙂 en.wikipedia.org/wiki/Bayesi…

New postdoc position in our group at @InfAtEd on RL and imitation learning for robot navigation in simulated environments using MIDGARD (midgardsim.org). Full position details: agents.inf.ed.ac.uk/Postdoc_… Application deadline: December 27th 2023

This week going to Boston to visit colleagues and give talks about our research and new MARL textbook (marl-book.com) at Northeastern, Harvard, MIT, UK consulate. Then next week to New Orleans for @NeurIPSConf! All welcome at talks, pls get in touch if you want to meet.