Alexey Skrynnik

Senior Research Scientist & Team Lead, AIRI

Moscow, Russia

Senior Research Scientist and Team Lead at AIRI focused on reinforcement learning, multi-agent systems, and LLM/multimodal agents. My recent work spans RL-style tree search for LLM test-time reasoning, foundation-model and learnable approaches to multi-agent pathfinding, and embodied instruction-following agents.

Research Experience

Senior Research Scientist & Team Lead

AIRI, Cognitive AI Systems Laboratory

Jul 2024 — Present Moscow, Russia
  • Leading an 8-person research team focused on RL, LLM/multimodal agents, and multi-agent systems
  • Supervised ReSCALE, an RL-style tree-search method for LLM test-time reasoning that restores monotonic scaling with larger search budgets without retraining (ICAPS 2026)
  • Led MAPF-GPT, a foundation model for multi-agent pathfinding with zero-shot generalization on unseen maps, outperforming state-of-the-art learnable solvers (AAAI 2025, Oral)

Research Scientist

AIRI, Cognitive AI Systems Laboratory

Aug 2021 — Jul 2024 Moscow, Russia
  • Learn to Follow (AAAI 2024, Oral): combined RL and decentralized planning for lifelong multi-agent pathfinding, improving generalization with a 10x speedup over a state-of-the-art search-based solver
  • Decentralized MCTS for partially observable MAPF: first MCTS approach for this setting (AAAI 2024)
  • Built and open-sourced POGEMA, a benchmark platform for multi-agent pathfinding, later published at ICLR 2025
  • RL Track Lead for the IGLU Competition at NeurIPS 2021 and 2022; co-developed benchmarks for collaborative embodied agents in grounded instruction-following Minecraft tasks for human-AI collaboration

Junior Research Scientist

Federal Research Center for Computer Science and Control, Russian Academy of Sciences

Feb 2018 — Aug 2021 Moscow, Russia
  • 1st place at the NeurIPS 2019 MineRL Diamond Competition with a hierarchical RL approach leveraging demonstrations as human priors for long-horizon decision-making in Minecraft; first author and presenter at NeurIPS
  • Researched multi-agent pathfinding, model-based RL, and visual navigation, including hybrid policy learning with classical search; published in IEEE TNNLS, Knowledge-Based Systems, and Cognitive Systems Research

Education

PhD in AI & Machine Learning FRC CSC RAS (defended at MIPT), Moscow, Russia, 2023
MS in Computer Science Rybinsk State Aviation Technical University, 2015 – 2017
BS in Computer Science Rybinsk State Aviation Technical University, 2011 – 2015

Selected Publications

ICAPS 2026

Revisiting Tree Search for LLMs: Gumbel and Sequential Halving for Budget-Scalable Reasoning

Leonid Ugadiarov , Yury Kuratov , Aleksandr Panov , Alexey Skrynnik

AAAI 2026 (Best Poster Award)

Camar: Continuous Actions Multi-Agent Routing

Artem Pshenitsyn , Aleksandr Panov , Alexey Skrynnik

IROS 2025

Advancing Learnable Multi-Agent Pathfinding Solvers with Active Fine-Tuning

Anton Andreychuk , Konstantin Yakovlev , Aleksandr Panov , Alexey Skrynnik

ICLR 2025

POGEMA: A Benchmark Platform for Cooperative Multi-Agent Pathfinding

Alexey Skrynnik , Anton Andreychuk , Artem Borzilov , Aleksandr Chernyavskiy , Konstantin Yakovlev , Aleksandr Panov

AAAI 2025 (Oral)

MAPF-GPT: Imitation Learning for Multi-Agent Pathfinding at Scale

Anton Andreychuk , Konstantin Yakovlev , Aleksandr Panov , Alexey Skrynnik

ACM SIGIR 2025

IDAT: A Multi-Modal Dataset and Toolkit for Building and Evaluating Interactive Task-Solving Agents

Shrestha Mohanty , Negar Arabzadeh , Andrea Tupini , Yuxuan Sun , Alexey Skrynnik , Artem Zholus , Marc-Alexandre Cote , Julia Kiseleva

ACL 2025

CrafText Benchmark: Advancing Instruction Following in Complex Multimodal Open-Ended World

Zoya Volovikova , Gregory Gorbov , Petr Kuderov , Aleksandr Panov , Alexey Skrynnik

ECAI 2024

Instruction Following with Goal-Conditioned Reinforcement Learning in Virtual Environments

Zoya Volovikova , Alexey Skrynnik , Petr Kuderov , Aleksandr I. Panov

AAAI 2024 (Oral)

Learn to Follow: Decentralized Lifelong Multi-Agent Pathfinding via Planning and Learning

Alexey Skrynnik , Anton Andreychuk , Maria Nesterova , Konstantin Yakovlev , Aleksandr Panov

AAAI 2024

Decentralized Monte Carlo Tree Search for Partially Observable Multi-Agent Pathfinding

Alexey Skrynnik , Anton Andreychuk , Konstantin Yakovlev , Aleksandr Panov

IEEE TNNLS

When to Switch: Planning and Learning for Partially Observable Multi-Agent Pathfinding

Alexey Skrynnik , Anton Andreychuk , Konstantin Yakovlev , Aleksandr Panov

NeurIPS 2022 Competition Track

Interactive Grounded Language Understanding in a Collaborative Environment: Retrospective on IGLU 2022 Competition

Julia Kiseleva , Alexey Skrynnik , Artem Zholus , Shrestha Mohanty , Negar Arabzadeh , Marc-Alexandre Cote , et al.

Technical Skills

Core Stack Python, PyTorch, JAX, C++
Research Areas Reinforcement Learning, Multi-Agent Systems, Planning/Search, Imitation Learning
LLM/Agent Systems RL for LLMs, LLM/Multimodal Agents, Test-Time Reasoning, Inference/Evaluation
Training/Infrastructure Distributed Training, FSDP/DDP, Slurm, Ray, vLLM, VERL, HF Transformers, LoRA/PEFT
Languages English (advanced), Russian (native)

Achievements & Service

Teaching

MSU AI Masters – Advanced Reinforcement Learning Lecturer (2025 – present)
MIPT – Reinforcement Learning, Software Tools for AI Assistant Lecturer (2020 – 2022)
HSE – Applied Problems of Data Analysis Lecturer (2018 – 2020)