news

Feb 12, 2026 ⭐️ MARL-GPT paper accepted to AAMAS-2026
Our paper MARL-GPT: Foundation Model for Multi-Agent Reinforcement Learning has been accepted as a full paper to AAMAS 2026 (AAAI Track). The conference will be held in person in Paphos, Cyprus in May 2026.
Jan 25, 2026 ⭐️⭐️⭐️⭐️ Four papers accepted to ICLR 2026
On January 25, 2026, the ICLR 2026 program committee confirmed acceptance of four of our submissions. All of them are now available on OpenReview:
  • ELMUR: External Layer Memory with Update/Rewrite for Long-Horizon RL – https://openreview.net/forum?id=bm3rbtEMFj
  • Memory, Benchmark & Robots: A Benchmark for Solving Complex Tasks with Reinforcement Learning – https://openreview.net/forum?id=9cLPurIZMj
  • Recurrent Action Transformer with Memory – https://openreview.net/forum?id=kByN4v0M3e
  • Unraveling the Complexity of Memory in RL Agents: an Approach for Classification and Evaluation – https://openreview.net/forum?id=lJKdOYFF5W
This result would not have been possible without my colleagues. I am sincerely grateful to the team for their collaboration, insightful discussions, criticism, ideas, and the enormous amount of time invested in these projects.
Jan 22, 2026 Memory Retention Is Not Enough has been submitted to arXiv
The camera-ready version of our AAMAS paper “Memory Retention Is Not Enough to Master Memory Tasks in Reinforcement Learning” is now available on arXiv.
Jan 21, 2026 KAGE-Bench preprint published
KAGE-Bench preprint is now available on arXiv! We propose an environment (and a benchmark built on it) for studying visual generalization in RL (i.e., training on one image distribution and validating on another). We wrote the benchmark in JAX because alternatives are too slow for rapid hypothesis testing. We also added the ability to customize the environment along 93 easily configurable parameters, as other benchmarks often confound multiple shifts in observation distributions (for example, changing not only illumination but also textures, etc.). Link to visualizations (there are many more in the paper): https://avanturist322.github.io/KAGEBench/ Paper: https://arxiv.org/abs/2601.14232 Code: https://github.com/CognitiveAISystems/kage-bench Hugging Face: https://huggingface.co/papers/2601.14232
Jan 13, 2026 KAGE-Bench code released
KAGE-Bench code is now available on GitHub! KAGE-Bench is a pure JAX 2D platformer RL benchmark for visual OOD generalization.
Dec 06, 2025 ⭐️ Memory-Rewrite has been accepted to AAMAS-2026
The paper “Memory Retention Is Not Enough to Master Memory Tasks in Reinforcement Learning” has been accepted to AAMAS-2026.
Dec 03, 2025 Announcing Awesome Memory VLA
We are releasing Awesome Memory VLA, a curated and continually updated collection of research on memory-augmented Visual Language Action models. The list focuses on methods for long-horizon, partially observable, and embodied decision making, bringing together recent work at the frontier of VLMs, control, and reinforcement learning. The repository tracks papers, benchmarks, datasets, and other resources relevant to building VLA systems that integrate memory for robust generalization and long-term reasoning. Community contributions are welcome. If you have a relevant paper or resource, feel free to submit a pull request.
Online
Dec 01, 2025 MARL-GPT has been accepted to WoMAPF @ AAAI-2026
The paper “MARL-GPT: Foundation Model for Multi-Agent Reinforcement Learning” has been accepted to WoMAPF Workshop @ AAAI-2026.
Nov 20 - Nov 21, 2025 Attending AIJ 2025
I participated in the AIJ-2025 Deep Dive track in a poster session with the papers “Memory, Benchmark & ​​Robots: A Benchmark for Solving Complex Tasks with Reinforcement Learning”, “MARL-GPT: Foundation Model for Multi-Agent Reinforcement Learning”, which I co-authored, and “Don’t Blind Your VLA: Aligning Visual Representations for OOD Generalization” by my amazing colleague Nikita Kachaev. I also gave a live talk on “Memory, Benchmark & ​​Robots”.
Offline, Moscow, Russia
Oct 24 - Oct 25, 2025 Attending Fall into ML 2025
I would like to thank my colleagues for presenting at Fall into ML 2025 the papers Recurrent Action Transformer with Memory and Memory, Benchark & ​​Robots: A Benchmark for Solving Complex Tasks with RL, which were accepted to NeurIPS-2025, including an oral presentation on the second paper. See posters page for details.
Offline, Moscow, Russia
Oct 09, 2025 ELMUR has been submitted to arXiv
The preprint of our paper “A New Perspective on Transformers in Online Reinforcement Learning for Continuous Control” is now available on arXiv.
Oct 09, 2025 ELMUR has been submitted to arXiv
The preprint of our paper “ELMUR: External Layer Memory with Update/Rewrite for Long-Horizon RL” is now available on arXiv.
Sep 27 - Sep 30, 2025 Attending CoRL-2025
From September 27 to September 30, 2025, I will be participating in the CoRL-2025 conference. Three papers will be presented at the RemembeRL workshop:
  1. ELMUR: External Layer Memory with Update/Rewrite for Long-Horizon RL (spotlight, best poster award)
  2. Memory, Benchmark & Robots: A Benchmark for Solving Complex Tasks with Reinforcement Learning (poster)
  3. Unraveling the Complexity of Memory in RL Agents: an Approach for Classification and Evaluation (poster)

Offline, Seoul, South Korea