Sheza Abbas Naqvi

Sheza Abbas Naqvi

Robotics Engineer | Autonomous UAVs & SLAM | Edge AI Optimization

Based in: Islamabad, PakistanCurrently: Teaching Assistant – AI for Robotics (AI 641 - Graduate Level), Lahore University of Management Sciences (LUMS)Experience: 1 year

Hire me

Robotics engineer and researcher focused on autonomous systems, SLAM, and edge-deployed AI. Proven track record in developing high-precision UAV-based phenotyping (97.6% IoU) and optimizing deep learning models for low-latency edge performance (70%+ reduction). Expert in ROS2, path planning, and multi-agent systems.

Experience

Teaching Assistant – AI for Robotics (AI 641 - Graduate Level)

Lahore University of Management Sciences (LUMS)Lahore, Pakistanpart time2026 – Present
Key Responsibilities: • Orchestrated the assessment framework for graduate-level implementations of path planning and SLAM. • Facilitated practical lab execution for complex robotic systems, providing technical support for ROS2-based stacks.
ROS2Path PlanningSLAM

Teaching Assistant – Junior Design Studio Robotics (EE3002)

Lahore University of Management Sciences (LUMS)Lahore, Pakistanwork2026 – May 2026

Summary

Redesigned physical lab modules centered on robotic manipulation, kinematics, and ROS integration. Guided incoming cohorts through hands-on hardware-software integration for mobile platforms.

Results

  • Independently designed and led 3 weeks of lab modules from scratch, focusing on robotic manipulation, kinematics, and ROS integration.
  • Mentored a cohort of 20 students, enabling them to successfully pursue advanced course projects in robotic manipulation.
  • Improved the lab ecosystem by introducing new equipment-handling protocols that increased student autonomy and project complexity.
  • Mentored 20 students.
  • Led 3 weeks of independent curriculum.
ROSRobotic ManipulationKinematics

Research Assistant - Autonomous UAV-Based 3D Phenotyping & Crop Mapping

LUMS Center for Water Informatics (WIT)Lahore, PakistanworkJun 2025 – May 2026

Summary

Executed a multi-month autonomous temporal monitoring campaign over a wheat field, deploying a custom UAV equipped with a Livox MID360 LiDAR, Arducam B0495 RGB camera, Pixhawk 6C, and Odroid N2+.

What I did

  • Field Campaign & Hardware: Executed a multi-month autonomous temporal monitoring campaign over a wheat field, deploying a custom UAV equipped with a Livox MID360 LiDAR, Arducam B0495 RGB camera, Pixhawk 6C, and Odroid N2+.
  • Robust LiDAR Mapping: Processed 13 temporal LiDAR point clouds using FAST-LIVO2, implementing RANSAC-based ground normalization and field-aware two-stage segmentation to consistently track plot topologies.
  • Multi-Date Structural Alignment: Developed a correspondence-guided planar registration and overlap-restricted height differencing pipeline, achieving a median shared-support IoU of 97.6% and tracking 98.7 cm of seasonal canopy growth.
  • Quantitative Validation: Validated LiDAR-derived canopy height against manual field observations, achieving an MAE of 5.42 cm, RMSE of 6.50 cm, and highly accurate tracking across 24 plot-date observations.
  • RGB Spectral Analysis: Generated RGB orthomosaics and automatically partitioned indexed plots to extract canopy vegetation indices (ExG, VARI, GLI) for multimodal temporal phenotyping.

Results

  • Achieved a median shared-support IoU of 97.6% in structural alignment.
  • Tracked 98.7 cm of seasonal canopy growth.
  • Validated LiDAR-derived canopy height with an MAE of 5.42 cm and RMSE of 6.50 cm across 24 plot-date observations.
C++PythonROS2FAST-LIVO2RANSACOpenCVArduPilotLivox MID360 LiDARArducam B0495 RGB cameraPixhawk 6COdroid N2+

Skills

Technical

PythonPython
MAVROS
Computer Vision
ROS/ROS2
Deep Learning
LiDAR-Inertial Fusion
Pixhawk 6C
C++C++
PyTorchPyTorch
TensorFlowTensorFlow
Multi-Agent Reinforcement Learning (MARL)
ArduPilot
FAST-LIVO2
RANSAC
GitGit
LinuxLinux
FAST-LIO
LIO-SAM
MAVLink
Odroid N2+
Raspberry PiRaspberry Pi
Livox Mid-360 LiDAR
Intel RealSense D455
OpenCVOpenCV
Open3D
PX4
UART/I2C/Serial
Keras
DockerDocker
MATLAB
Agisoft Metashape
SQLSQL
Hugo
Topometric Mapping
Stochastic Scheduling

Languages

English

Projects

Visual-to-Motion Robotic Drawing using ROS and OpenCV

Developeracademic
Developed a vision-guided sketching system enabling an Interbotix WidowX 250s arm to execute dynamic drawing tasks via MoveIt. Performed Hand-Eye extrinsic calibration between a RealSense D455 and the robotic arm using projective geometry to map visual features to the robot’s coordinate frame.
ROSMoveItOpenCVPythonRVizRealSense D455Interbotix WidowX 250s

Robust Real-Time Deep Learning Prediction on Edge Devices

Lead Developerpersonal
Developed a dual-head ResNet-based deep learning system for simultaneous regression and classification tasks. Integrated Bayesian inference with Monte Carlo dropout to quantify predictive uncertainty for robust edge deployment, managing noisy data streams. Optimized model for in-browser deployment via TensorFlow.js, enabling real-time, low-latency inference.
TensorFlowKerasPythonResNetNeural NetworksTensorFlow.jsBayesian inference

Stochastic Utility-Aware Multi-Robot Task Allocation

Developerpersonal
Designed a warehouse coordination system combining Stochastic Utility Allocation (SCOBA) with Priority-Based Conflict-Based Search (PM-CBS). Implemented EKF localization on topometric region graphs to robustly track trajectories under sensing noise. Achieved zero collisions and superior utility efficiency compared to Greedy A* and Q-learning baselines.
PythonPath Planning AlgorithmsEKFSimulationSCoBAPM-CBSTopometric Mapping

Safe Multi-Agent RL for Cooperative Navigation

Developeracademic
Developed a decentralized navigation framework combining Soft Actor-Critic (SAC) with Model Predictive Control (MPC) safety shields to guarantee constraint satisfaction. Synthesized a custom Python simulation environment to train agents in obstacle-dense scenarios, achieving an 89.4% success rate. Implemented a three-stage curriculum learning strategy for progressive obstacle reasoning.
PythonPyTorchACADOS (MPC)NumPy

Robust Real-Time Age & Gender Prediction on Edge Devices

Developeracademic
Developed a dual-head ResNet-based deep learning system for simultaneous age regression and gender classification. Integrated Bayesian inference with Monte Carlo dropout to quantify predictive uncertainty for robust edge deployment. Optimized model for in-browser deployment via TensorFlow.js, enabling real-time, low-latency inference.
TensorFlowKerasPythonResNetTensorFlow.js

Autonomous UAV-Based Crop Growth Mapping - Senior Year Project

Lead Researcher/Engineerpersonal
Hardware Architecture: Deployed a custom 500mm carbon-fiber quadrotor, integrating a Livox Mid-360 LiDAR and Odroid N2+ payload. Designed and troubleshooted custom electrical power distribution systems and wiring schematics for high-heat field operations. Control System: Engineered a MAVROS-based state estimation bridge to provide high-frequency local position feedback ("Virtual GPS") to the Pixhawk 6C controller, enabling stable autonomous flight in GPS-denied environments. Validation & Ground Truth: Designed an extrinsic calibration system using AprilTags to quantify SLAM drift, serving as a field-deployable proxy for Motion Capture ground truth. Embedded Optimization: Ported and optimized the FAST-LIVO2 stack for ARM64 (Odroid) architecture, managing thread allocation to guarantee real-time 50Hz state estimation performance. Digital Twin Generation: Constructed high-fidelity 4D spatiotemporal digital twins by processing multi-temporal datasets through Agisoft Metashape and Open3D for granular volumetric analysis of crop health.
C++PythonROS2FAST-LIVO2ArduPilotOpenCVLinuxHardware-in-the-Loop (HIL)Agisoft MetashapeOpen3DPixhawk 6CLivox Mid-360 LiDAROdroid N2+

Education

Bachelor of Science

Lahore University of Management Sciences (LUMS)Electrical Engineering | Minor in Computer ScienceSep 2022 – May 2026

Thesis. Autonomous UAV-Based Crop Growth Mapping

Coursework

Mobile RoboticsJunior Design Studio (Robotics)Dynamic Programming and Reinforcement LearningMachine LearningComputer VisionDeep LearningCommunication SystemsFeedback Control SystemsSignals and Systems

Recognition

Certifications

TOEFL iBT

ETS

TOEFL iBT (Score: 110)

ETS

What drives the work

Motivations

  • Robotics Engineering
  • Autonomous UAVs
  • Computer Vision
  • SLAM Systems
  • Precision Agriculture
  • Reinforcement Learning
  • Multi-Agent Systems
  • Path Planning
  • AgTech Startups
  • LiDAR Fusion
  • Research & Development
Hire ShezaReach out about a role, a contract or a conversation.For recruiters

Sheza's twin is AI, it can make mistakes.

More people on Kanvis

Browse more peopleCreate your own Kanvis