Omer Tariq
Email Scholar GitHub LinkedIn CV

Omer Tariq

Research Scientist — 3D Perception & Embodied AI
Neubility, Seoul, South Korea

I am a Research Scientist at Neubility, building multi-camera 3D perception and Vision-Language-Action systems for autonomous mobile robots. I hold a Ph.D. in Computer Science from KAIST (2025), advised by Prof. Dongsoo Han.

My research focuses on transformer-based and BEV architectures for multi-camera 3D object detection, VLM/VLA model deployment on NVIDIA Jetson Orin via TensorRT INT8/FP16 quantization, and sim-to-real transfer for autonomous robot perception. At Neubility I deliver production perception pipelines running sub-30 ms end-to-end on a single SoC — achieving 8× inference speedup at <1% mAP degradation.

I am interested in research roles at the intersection of 3D scene understanding, embodied AI, and efficient edge deployment.

KAIST Ph.D. CS — KAIST (2025) UET B.S. EE — UET Taxila (2014)

Recent News

Jan 2026 Paper accepted to IJCNN 2026 (Maastricht, Netherlands): "Uncertainty Aware and Decoder Aligned Learning for Video Summarization"
Aug 2025 Joined Neubility (Seoul) as a Research Scientist, working on multi-camera 3D perception and Embodied AI for autonomous delivery robots
Jul 2025 Thrilled to announce my latest publication in the IEEE Internet of Things Journal (Impact Factor: 8.9)! 🚀 🔬 NanoMST: A Hardware-Aware Multiscale Transformer Network for TinyML-Based Real-Time Inertial Motion Tracking
Jun 2025 Successfully defended my Ph.D. dissertation at KAIST — Thesis: Robust Domain-Invariant Inertial Localization in Real-Time Federated Edge Computing
Mar 2025 Started as a Machine Learning Engineer at Monitra, UK
Mar 2025 Started as a Teaching Assistant for Special Topics in Computer Science (CS492) at KAIST

Selected Publications

Full list on Google Scholar  ·  All publications

NanoMST: A Hardware-Aware Multiscale Transformer Network for TinyML-Based Real-Time Inertial Motion Tracking
Omer Tariq, Dongsoo Han
IEEE Internet of Things Journal, 2025
NanoMST is an efficient multi-scale transformer for inertial motion tracking that achieves high-precision trajectory estimation with minimal computational cost and real-time performance on edge devices, outperforming larger models across benchmarks while requiring only 298K parameters and supporting 8-bit quantization.
Paper
DeepILS: Towards Accurate Domain Invariant AIoT-enabled Inertial Localization System
Omer Tariq, B. Dastagir, M. Bilal, Dongsoo Han
IEEE Internet of Things Journal, 2025
This paper presents DeepILS, a domain-invariant AIoT-enabled inertial localization system that achieves high accuracy across diverse environments without requiring environment-specific retraining.
Paper Code
Meta-Swin: Lightweight Image Super-Resolution Swin Transformer for Metaverse Applications
Omer Tariq, Dastagir M.B.A., M. Bilal, Dongsoo Han
2nd International Conference on Intelligent Metaverse Technologies & Applications (iMETA), 2024
This paper proposes Meta-Swin, a lightweight Swin Transformer architecture for image super-resolution in metaverse environments with constrained computational resources.
Paper
EmoHEAL: A Fusion-Based Framework for Emotion Recognition Using Wearable Sensors
Omer Tariq, Y. Oh, Dongsoo Han
2024 IEEE SENSORS, 2024
EmoHEAL is a fusion-based framework that leverages multimodal wearable sensor data for accurate and robust emotion recognition with applications in health monitoring.
Paper Code

View all 15+ publications →

Honors & Awards

  • KAIST MS/Ph.D. Full Scholarship (20,000 USD/year) 2021
  • Distinguished Service Award, SUPARCO 2018
  • Distinguished Performance Award — PakTes-1A satellite launch 2018

Academic Service

Reviewer

  • IEEE Internet of Things Journal
  • IEEE Internet of Things Letters
  • Engineering Applications of AI
  • Nuclear Science & Techniques