CV

Download CV (PDF)

Summary

Ph.D. Candidate in Computer Science with 4+ years of Army Research Laboratory (ARL)-funded research in GPS-denied visual localization, multi-sensor fusion, and deep learning for autonomous navigation. Published IEEE author (3 peer-reviewed venues) with end-to-end experience from algorithm prototyping to field validation.

Education

Ph.D. Candidate, Computer Science | GPA: 3.7/4.0 Missouri University of Science and Technology, Rolla, MO | Aug 2021 – Present

• Advisor: Dr. Sanjay Madria (Army Research Laboratory funded)
• Focus: GPS-Denied Visual Localization, Deep Learning, Multi-Camera Stereo Systems, Sensor Fusion
• Relevant Coursework: Advanced ML in Computer Vision, Machine Learning, Data Mining, NLP

Research Experience

Graduate Research Assistant – Army Research Laboratory Funded Missouri University of Science and Technology, Rolla, MO | Aug 2021 – Present

• GPS-Denied Localization Framework (LanBLoc): Designed and implemented a stereo-vision-based absolute localization system for GNSS-denied outdoor environments; integrated YOLO-based object detection with stereo ranging for landmark-referenced positioning.
• Navigation Pipeline: Developed vision-driven navigation pipelines coupling visual localization with obstacle avoidance and path optimization for autonomous ground platforms in GPS-degraded scenarios.
• Sensor Fusion & Multi-Metric Localization: Built an image-retrieval-based localization method using dynamic vicinity clustering and multi-metric descriptor scoring (centroid similarity, PCA reconstruction error, k-NN matching) to handle geometric ambiguity in landmark-sparse environments.
• SLAM/VO Evaluation: Installed, configured, and benchmarked multiple SLAM and visual odometry systems (DROID-SLAM, DPV-SLAM, GO-SLAM, DPVO, ORB-SLAM3) on real stereo datasets; developed evaluation pipelines for trajectory accuracy (ATE/RTE metrics).
• Cross-View Geo-Localization (HCVGLoc): Engineered a multi-GPU DDP training pipeline in PyTorch for UAV-to-satellite geo-localization, incorporating domain adversarial training, rotation-aware modules, and uncertainty quantification for GPS-denied UAV navigation.
• Dataset Collection & Validation: Built and released a real-world stereo landmark dataset; designed end-to-end data collection, annotation, and validation pipelines for large-scale outdoor image datasets.
• Authored 3 peer-reviewed IEEE publications (WoWMoM 2025, WoWMoM 2024, AIPR 2023); active peer reviewer for IEEE conferences and journals.

Professional Experience

Graduate Teaching Assistant Missouri University of Science and Technology, Rolla, MO | Aug 2024 – Present

• Led Data Structures lab sessions in C++ and delivered hands-on big data technology sessions (Hadoop MapReduce, Spark, HBase, MongoDB).

Lecturer, Computer Science Sagarmatha Engineering College, NP | Feb 2019 – Aug 2021

• Taught programming, databases, and operating systems; supervised undergraduate ML/CV capstone projects.

Software Engineer Pioneer Solutions LLC (acquired by Hitachi Energy), Denver, CO | Dec 2017 – Feb 2019

• Developed and maintained ERP modules using SQL Server, PHP, and JavaScript.

Skills

• Languages: Python, C/C++, Bash, R, JavaScript, SQL
• Navigation & Perception: Visual SLAM/VO (ORB-SLAM3, DROID-SLAM, DPV-SLAM, GO-SLAM, DPVO), Stereo Vision, Structure-from-Motion (SfM), Multi-View Stereo (MVS), 3D Reconstruction
• Sensor Fusion: Extended Kalman Filter (EKF), Multi-Sensor Fusion (stereo cameras, IMUs), Sensor-Agnostic Localization
• Deep Learning & CV: PyTorch, TensorFlow, OpenCV, MMCV, Object Detection (YOLO, DETR, SSD, RCNN), Transformers, CNNs
• Simulation & Tools: ROS (familiarity), Git/GitHub, Docker, Linux/UNIX, Jupyter, NumPy, Pandas, Matplotlib
• Hardware Familiarity: NVIDIA GPUs (multi-GPU DDP training), Stereo Camera Rigs, Edge Deployment Pipelines

Selected Publications

1. Sapkota, G., & Madria, S. (2025). SafeNav: Safe Path Navigation using Landmark Based Localization in GPS-denied Environment. IEEE WoWMoM 2025.
2. Sapkota, G., & Madria, S. (2024). Landmark-based Localization using Stereo Vision and Deep Learning in GPS-Denied Battlefield Environment. IEEE WoWMoM 2024.
3. Sapkota, G., & Madria, S. (2023). Landmark Stereo Dataset for Landmark Recognition and Moving Node Localization. IEEE AIPR 2023.

Selected Projects

HCVGLoc – Hierarchical Cross-View Geo-Localization for UAVs

• Designed a 3-stage hierarchical localization pipeline for GPS-denied UAV navigation using satellite imagery, incorporating domain adversarial learning, uncertainty quantification, and Extended Kalman Filter (EKF) fusion.
• Trained on multi-GPU DDP setup (4x GPU) with datasets CVUSA, CVACT, VIGOR, and University-1652; implemented gallery-based evaluation with Recall@K metrics.

Radiogenomic Brain Tumor Classification

• Deep learning pipeline for MGMT methylation prediction from MRI integrating DenseNet-169 and SAM; achieved 70% accuracy and 67.54% AUC (RSNA-MICCAI).

Honors & Activities

• Best Poster Award (1st Place) – ISC Poster Presentation, Missouri S&T, Nov 2024
• Best Poster Award (1st Place) – Pathways Symposium, NextGen Precision Health 2026
• Founder & President, Nepali Students’ Association (NSA-MST), Aug 2023 – Present
• Finance Director, Council of Graduate Students (CGS), Missouri S&T, Aug 2024 – Present
• Certifications: Scientific Computing with Python; ML & Statistical Analysis (WorldQuant University)