Pavel Kopanev

Hi, I’m Pavel, an ML & 3D Computer Vision Research Engineer at the Technology Innovation Institute, developing Embodied AI for robotic manipulation, currently on humanoid platforms.

My current focus spans Vision-Language-Action models, RGB-D and point-cloud perception, 6D object pose estimation, synthetic data, and real-time deployment on robotic hardware. Previously, I was a Computer Vision Research Engineer at Huawei, contributing to multi-modal 3D object detection for autonomous driving. Before that, I joined a small startup building tools to make neural rendering more user-friendly and reconstruct 3D from smartphone photos.

Earlier, I contributed to visual-inertial SLAM systems at Integrant and spent a short time as a Research Intern in the 3D Deep Learning group at Skoltech.

Research interests

embodied AI VLA models 3D computer vision 6D pose estimation SLAM generative AI foundation models synthetic data neural rendering robot learning

CV Scholar GitHub LinkedIn X

Away from the screen I travel as much as I can and take photos along the way. Here’s a small collection. And if you’re into robots, .

Activity

Latest from X

I use X mostly as a lightweight research notebook: papers, benchmarks, datasets, robotics demos, and reposts around 3D perception, VLA models, embodied AI, and CV/robotics systems.

Open @kopanevp on X →

Loading latest post…

Experience

2022–present

Technology Innovation Institute (TII)

Computer Vision Engineer / Lead Embodied AI Engineer · Abu Dhabi

Leading a small embodied-AI team building VLA models for single-arm, dual-arm, mobile-manipulation, and humanoid platforms.
Working end-to-end: robot data collection, pre-training, flow-based action learning, task fine-tuning, evaluation, and deployment on real hardware.
Building RGB-D perception and manipulation pipelines with 6D object pose, grasp sampling, motion planning, collision checking, and ROS2 integration.
Using synthetic-data tools such as Gaussian Splatting, Blender, Unity, and custom 3D reconstruction pipelines for robotics datasets.

2022

Huawei

Computer Vision Research Engineer · AI Foundation and Algorithm Lab

Worked on multimodal 3D object detection for autonomous driving, combining LiDAR, camera, and radar signals.
Focused on perception model development, experiments, and evaluation for driving scenes.

2021–2022

Teleport

3D Computer Vision Engineer · Startup

Built pieces of a production-oriented 3D reconstruction pipeline around photogrammetry, neural rendering, and capture workflows.
Worked on real-time neural scene rendering and photo-to-video generation with inpainting, segmentation, and monocular depth estimation.
Explored single-view and sparse-view 3D reconstruction methods for faster 3D content creation.

teleport3.io

2021–2022

Skolkovo Institute of Science and Technology

Research Intern · ADASE 3DDL

Built visualization tools for large 3D models and point clouds.
Contributed to multi-sensor dataset collection for multi-view 3D reconstruction, later published at CVPR 2023.

CVPR 2023 paper

2021

Integrant

3D Computer Vision Engineer · SLAM Group

Contributed to visual and visual-inertial SLAM pipelines, with a focus on tightly coupled IMU integration into a visual SLAM framework.

2020

Sberbank Robotics Laboratory

Robotics Research Intern

Contributed to a systematic review of modern visual SLAM algorithms, later published in the Journal of Intelligent and Robotic Systems.

JIRS paper

Publications

Selected papers from my CV. Full publication list is on Google Scholar.

UR5 manipulation setup and trajectory plot from the Keypoint Priors paper — Robot demo + trajectory prior

CoRL Workshop2025

Generalization of Manipulation Skills using Keypoint Priors

Authors

G. Franzese, D. Sotelo, M. O. Elnour, P. Kopanev, A. Rehman, M. Asokan, D. Caporale, P. Ardón

Venue

CoRL 2025 Workshop on Generalizable Priors for Robot Manipulation

Description

Sparse 3D keypoint priors for making manipulation skills generalize better across new scenes and objects.

PDF

Rendered 3D objects from the T3D reconstruction paper — 3D reconstruction system

GraphiCon2025

T3D: Modular hardware and software complex for reconstruction of 3D objects

Authors

A. O. Zhirkov, R. O. Rodionov, P. D. Kopanev, V. A. Frolov

Venue

GraphiCon 2025 · pp. 133–145

Description

A modular hardware/software setup for reconstructing objects and turning captures into usable 3D assets.

PDF

CVPR2023

Multi-Sensor Large-Scale Dataset for Multi-View 3D Reconstruction

Authors

O. Voynov, G. Bobrovskikh, P. Karpyshev, S. Galochkin, A.-T. Ardelean, A. Bozhenko, E. Karmanova, P. Kopanev, Y. Labutin-Rymsho, R. Rakhimov, A. Safin, V. Serpiva, A. Artemov, E. Burnaev, D. Tsetserukou, D. Zorin

Venue

CVPR · pp. 21392–21403

Description

A multi-sensor benchmark with RGB, depth, infrared, lighting variation, and challenging materials for 3D reconstruction.

PDF

Visual SLAM trajectory and GUI visualization from the Visual SLAM comparison paper — SLAM benchmark view

JIRS2023

Comparison of modern open-source Visual SLAM approaches

Authors

D. Sharafutdinov, M. Griguletskii, P. Kopanev, M. Kurenkov, G. Ferrer, A. Burkov, A. Gonnochenko, D. Tsetserukou

Venue

Journal of Intelligent and Robotic Systems · 107(3), 43

Description

A practical comparison of open-source visual SLAM systems across accuracy, speed, robustness, and usability.

Springer · arXiv

DeltaCharger prototype photo from the RA-L paper — Charging robot prototype

RA-L2021

DeltaCharger: Charging Robot With Inverted Delta Mechanism and CNN-Driven High Fidelity Tactile Perception for Precise 3D Positioning

Authors

I. Okunevich, D. Trinitatova, Pavel Kopanev, D. Tsetserukou

Venue

IEEE Robotics and Automation Letters · 6(4), 7604–7610

Description

A charging robot with an inverted delta mechanism and tactile perception for precise electrode alignment.

IEEE Xplore · arXiv

Education

Skolkovo Institute of Science and Technology

M.Sc. in Information Systems and Technology · GPA 5.0/5.0

2019–2021 · Moscow

M.Sc.
thesis

Visual-Inertial SLAM for an autonomous outdoor delivery robot

Tightly coupled wheel odometry, visual-inertial localization, and a DNN-based front-end for robust outdoor robot navigation.

Localization package · ★ – ⑂ –

Bauman Moscow State Technical University

B.Sc. in Robotics and Mechatronics · GPA 5.0/5.0

2015–2019 · Moscow

B.Sc.
thesis

Computer vision for autonomous ROV docking

Vision components for automating underwater robot docking with a station.

Selected Robotics Projects

A couple of student-team projects from my university days.

Skoltech Eurobot Reset team at a robotics competition

Reset · Skoltech Eurobot Team

Software Developer / Team Mentor · 2019–2021

EUROBOT is a student robotics competition I took part in during my first year at Skoltech. I worked on localization and navigation for two autonomous robots, and later helped mentor the team. In 2021 the team went on to win the Russian final stage of EUROBOT.

Hydranautics team celebrating at SAUVC 2019 with an autonomous underwater robot

Hydranautics · Bauman BMSTU Team

Software Developer · 2016–2020

Hydranautics is where I got my first serious taste of underwater robotics. I worked on computer vision, mission logic, simulation, and the control GUI for ROV/AUV systems. The team finished 4th at SAUVC 2019 with the first AUV design built in the lab.

Small open-source projects

Not much, unfortunately; most of my work lives in private repos.

x-reposts-obsidian

Loading…

★–stars ⑂–forks ◉–watchers

First commit – · Last commit –

SLAM-Dockers

Loading…

★–stars ⑂–forks ◉–watchers

First commit – · Last commit –

SuperGlue_ROS

Loading…

★–stars ⑂–forks ◉–watchers

First commit – · Last commit –

Extracurricular Activities & Certificates

Control. Information. Optimization

Aug 2020

Sirius University · Sochi, Russia

Scientific and educational school-conference focused on optimal control, discrete and continuous optimization, statistics, machine learning, and mathematical modeling.

Certificate

Summer School of Machine Learning

Aug 2020

Skolkovo Institute of Science and Technology · Moscow, Russia

A one-week intensive online course on modern statistical machine learning. Presented the project Exploring Autoencoders and Contrastive Learning in Application to Deep Reinforcement Learning.

GitHub · Certificate