Nikita Kachaev

Hi there! I’m a Research Engineer developing new methods for embodied AI, robot learning, and Vision-Language-Action (VLA) models. I split my time between research and engineering: building, evaluating, and deploying models for robotic systems, while writing papers at the intersection of online reinforcement learning, foundation models, and robotics.

Google Scholar     X (Twitter)     LinkedIn     Github     Telegram     Views: Loading

ttonyalpha at gmail dot com

Nikita Kachaev

Highlights & News

🎉 Two papers accepted to the EMNLP 2026 main track: Does VLA Even Know the Basics? (Act2Answer) and VLA Grounder!

🎙️ Featured in the RLWRLD interview with AI researchers at ICML 2026.

🎉 Two ICLR 2026 posters: MIKASA-Robo and Unraveling the Complexity of Memory in RL Agents.

🎉 Don’t Blind Your VLA accepted at AAMAS 2026 as an oral presentation.

📄 Mind and Motion Aligned published in Reports of the RAS: Mathematics, Informatics, and Control Processes.

EMNLP 2026, Main

Does VLA Even Know the Basics? Limitations in Commonsense and World Knowledge of VLA Models

Nikita Kachaev, Andrey Moskalenko, Matvey Skripkin, Nikita Kurlaev, Daria Pugacheva, Albina Burlova, Mikhail Kolosov, Denis Shepelev, Andrey Kuznetsov, Elena Tutubalina, Aleksandr Panov, Alexey Kovalev, Vlad Shakhuro
EMNLP 2026, Main
Project Page  •  PDF

Act2Answer evaluates commonsense and world knowledge through robotic actions, revealing limitations in VLA models.

EMNLP 2026, Main

VLA Grounder: Language-Conditioning Space Optimization for Black-Box VLA Models

Damir Shodiev, Nikita Kachaev, Alexey Kovalev, Aleksandr Panov, Aleksei Staroverov
EMNLP 2026, Main
Project Page  •  Paper

A reasoning VLM policy trained with GRPO and sparse rewards optimizes instructions for frozen, black-box VLA models.

AAMAS 2026, Oral

Don’t Blind Your VLA: Aligning Visual Representations for OOD Generalization

Nikita Kachaev, Mikhail Kolosov, Daniil Zelezetsky, Alexey K. Kovalev, Aleksandr I. Panov
AAMAS 2026, Oral
Project Page  •  PDF  •  Code

Visual representation alignment improves real-robot success rates by approximately 30% in previously unseen environments.

under review

μVLA: On Recurrent Memory for Partially Observable Manipulation in VLA Models

Egor Cherepanov, Nikita Kachaev, Daniil Zelezetsky, Aydar Bulatov, Artem Pshenitsyn, Yuri Kuratov, Alexey Skrynnik, Aleksandr I. Panov, Alexey K. Kovalev
under review
Project Page  •  Code

Investigating recurrent memory for VLA manipulation under partial observability.

ICLR 2026, Poster

Memory, Benchmark & Robots: A Benchmark for Solving Complex Tasks with Reinforcement Learning

Egor Cherepanov, Nikita Kachaev, Alexey K. Kovalev, Aleksandr I. Panov
International Conference on Learning Representations (ICLR 2026), Poster
Project Page  •  PDF  •  Code

MIKASA-Robo benchmarks memory-intensive robotic tasks for reinforcement learning agents.

ICLR 2026, Poster

Unraveling the Complexity of Memory in RL Agents: an Approach for Classification and Evaluation

Egor Cherepanov, Nikita Kachaev, Artem Zholus, Alexey K. Kovalev, Aleksandr I. Panov
International Conference on Learning Representations (ICLR 2026), Poster
PDF

A framework for classifying and evaluating the memory demands of reinforcement learning tasks and agents.

ICLR@WRL 2025

A New Perspective on Transformers in Online Reinforcement Learning for Continuous Control

Nikita Kachaev, Daniil Zelezetsky, Egor Cherepanov, Alexey K. Kovelev, Aleksandr I. Panov
The Thirteenth International Conference on Learning Representations (ICLR), 7th Robot Learning Workshop. Poster, 2025
PDF

RAS 2025, Journal

Mind and Motion Aligned: A Joint Evaluation Benchmark for Planning and Mobile Manipulation

Nikita Kachaev, Andrei Spiridonov, Andrey Gorodetsky, Kirill Muravyev, Nikita Oskolkov, Aditya Narendra, Vlad Shakhuro, Dmitry Makarov, Aleksandr I. Panov, Polina Fedotova, Alexey K. Kovalev
Reports of the RAS: Mathematics, Informatics, and Control Processes, Journal, 2025
PDF

A ROS2-compatible real2sim framework in NVIDIA Isaac Sim for evaluating planning, navigation, and mobile manipulation.

AAMAS 2026 oral presentation

⭐ Oral talk at AAMAS 2026

27 May 2026
Our work Don’t Blind Your VLA was presented as an oral talk at AAMAS 2026 in Cyprus: Daniil presented in person, while I joined online.

ICLR 2026 in Rio de Janeiro

🇧🇷 Presented 2 papers at ICLR 2026 in Rio de Janeiro, Brazil

22-27 April 2026
In Rio de Janeiro, Brazil I presented two works at ICLR 2026: Memory in RL Agents (📍Thursday, April 23, Poster P3-1721) and MIKASA (📍Saturday, April 25, Poster P4-5208).

AI in Robotics course

Invited lecture on VLA models for the YSDA AI in Robotics course

30 April 2026
I was invited to the YSDA AI in Robotics course to give a lecture and seminar on VLA models and modern robotic manipulation.

Seminar on VLA generalization and representation alignment

Seminar on VLA generalization and text-visual representation alignment

21 May 2026
I gave Seminar 16: Generalization of VLA models and text-visual representation alignment in VLA models, discussing why fine-tuned VLA models can forget, lose focus, and degrade language understanding on small robotics datasets, and how visual representation alignment helps mitigate these issues.

Workshop on VLA models

Organized workshop on VLA models at ROS Meetup 2026

21 March 2026
I organized and delivered a workshop on Visual-Language-Action models at ROS Meetup 2026, covering modern VLA training, evaluation, and practical robotics development workflows.

Transformer Models in Deep Learning course

Lecturer for Transformer Models in Deep Learning at HSE University

12 January 2026
I taught the Transformer Models in Deep Learning module course on attention, transformers, LLMs, VLMs, efficient inference, and applications across NLP, computer vision, and other domains.

Spotlight talk on memory in RL agents

Spotlight talk on Memory in RL Agents at HSE ML and AI Conference

12 December 2025
I gave a spotlight talk on memory in reinforcement learning agents at the HSE University 4th Conference on Machine Learning and AI.

Representation alignment invited talk

Invited talk on representation alignment for OOD generalization in VLA

14 November 2025
I gave an invited talk at the ITMO University Reading Club Special Session on representation alignment for out-of-distribution generalization in VLA models.

NVIDIA Isaac Sim workshop

Organized workshop on NVIDIA Isaac Sim at ROS Meetup 2025

10 October 2025
I organized and delivered a deep-dive workshop on NVIDIA Isaac Sim at ROS Meetup 2025, covering real2sim workflows and robotics simulation tooling.

Transformer-based online RL seminar

Seminar on transformer-based online reinforcement learning

12 December 2025
I presented transformer-based approaches for online reinforcement learning at AIRI and MIPT reading clubs and seminars.

AIRI Summer School VLA projects

Mentored student projects on VLA models at AIRI 2025 Summer School

18 August 2025
I mentored student projects on Visual-Language-Action models, reinforcement learning, and reasoning for embodied AI at AIRI 2025 Summer School.

Some random stuff will be here in the future.