This project presents a complete workflow for 3D scene reconstruction from aerial imagery captured using a monocular RGB camera mounted on the VOXL 2 drone platform. The process begins with carefully planned drone flights and synchronized data acquisition using ROS2, ensuring comprehensive coverage and accurate timestamping of both images and telemetry. To optimize reconstruction, a custom Python script employing ORB and FLANN selects frames that maximize viewpoint diversity while reducing redundancy. The curated images are processed through the open-source COLMAP pipeline, which performs feature extraction, structure-from-motion, and dense multi-view stereo reconstruction to produce detailed point clouds and textured meshes. Open3D is used for further point cloud processing and visualization. The pipeline’s performance was benchmarked against leading commercial photogrammetry solutions, PolyCam and Reality Capture, using the same datasets. Results showed that the open-source approach could generate dense, accurate 3D models, with RealityCapture offering the highest fidelity and fastest processing times. Overall, this project demonstrates that high-quality, reproducible 3D reconstruction is achievable with affordable hardware and open-source tools, supporting applications in mapping, inspection, and virtual modeling. 

Video 1 : COMAP Reconstruction

Video 2 : Polycam Reconstruction

Video 3 : Reality Capture Reconstruction

This project demonstrates a fully autonomous UAV system capable of detecting and tracking AprilTags using ROS 2 and PX4 offboard control. By leveraging pose data from a vision-based AprilTag detector, the drone computes spatial transforms in real time and generates precise trajectory setpoints to follow the tag dynamically. The system integrates tightly with PX4 flight control and TF2 transforms, enabling robust localization and control in real-world environments. This work showcases the potential of combining perception and control for intelligent aerial robotics applications.

This project explores the use of deep learning and eye-tracking data to predict Autism Spectrum Disorder (ASD) in children. By transforming temporal eye movement data into spatially meaningful images, the study applies a series of models—including custom CNNs, transfer learning, LSTM with attention, and Transformers—for both image-level and individual-level binary classification. The models are evaluated using medical-context metrics such as accuracy, AUC, and F1 score, with the Transformer-based model achieving the best performance (AUC of 0.95). This approach enables early, non-invasive ASD screening and supports clinical decision-making by modeling gaze behavior patterns over time.

This project showcases autonomous navigation using SLAM and action-client nodes in ROS2. Waypoints are generated based on the detected positions of batteries placed within the environment, as captured by cameras. The robot's current position and waypoint locations are determined by referencing the TF tree for precise navigation.

An autonomous robot was developed to improve waste management in cafeterias by detecting, collecting, and disposing of discarded aluminum cans. Built using ROS2 Humble and simulated in Gazebo, the system leverages the Nav2 stack for real-time navigation and obstacle avoidance, while OpenCV enables image-based can detection under randomized positioning. Developed in modern C++17, the system employs test-driven development with robust unit and integration testing using Colcon. Continuous integration and deployment are managed via GitHub Actions, incorporating automated workflows for testing, code coverage (Codecov), and static code analysis with Clang-Tidy and CPPLint. Agile methodologies guided iterative development, with structured Git branching strategies maintaining code integrity. The solution reduces labor costs and promotes hygienic recycling while demonstrating scalable robotic automation through advanced software engineering practices. 

This project implements autonomous navigation and obstacle avoidance using Deep Q-Networks (DQN) on the TurtleBot3 platform. Leveraging LIDAR and odometry data, the robot learns to navigate through static and dynamic obstacles in a simulated Gazebo environment, optimizing its movement toward goal positions. The system integrates ROS2 for communication and control, and employs a custom-designed deep neural network for policy learning. The DQN model is trained using a reward-driven framework to improve navigation accuracy and decision-making over time. The simulation features randomized goals and obstacles to enable adaptive learning under realistic, dynamic conditions, demonstrating the effectiveness of reinforcement learning in real-world robotic navigation tasks. 

This project involves developing a ROS package that enables a TurtleBot to autonomously navigate through a maze using Aruco markers for localization. As the robot traverses the maze, it detects and reports objects present in the environment, integrating perception and navigation capabilities to achieve efficient maze traversal. 

Conducted kinematic analysis and Gazebo-based simulation of the ABB IRB 1600 industrial robot for CNC tending applications, covering forward and inverse kinematics and validating robotic motion in a virtual environment. Demonstrated high-performance cycle times and precision in machine tending operations. 

MULTI⦿VIZ is a graphical user interface (GUI) designed to enhance multi-robot coordination and visualization. Built using the ROS2 framework, it provides real-time monitoring, teleoperation, and control of multiple robots with features such as voice commands, emergency stop functionality, and sensor data visualization (LiDAR, odometry, and camera feeds). The tool addresses challenges in managing multi-robot systems by offering an intuitive and scalable interface for research and practical deployment. Developed as an open-source project, MULTI⦿VIZ aims to promote collaboration and accessibility within the robotics community. This project is still under development.

This project focuses on implementing the Real-Time Rapidly-exploring Random Tree Star (RT-RRT*) algorithm to enhance mobile robot navigation in dynamic environments. RT-RRT* is an extension of the RRT* algorithm, designed to efficiently update paths in response to changes, enabling robots to adapt their trajectories in real-time as obstacles move or new information becomes available. This approach is particularly beneficial for applications requiring continuous path optimization in unpredictable settings.

This project focuses on implementing the A* algorithm for path planning in a differential drive (non-holonomic) robot. The A* algorithm is utilized to compute an optimal path from a start to a goal position, considering the robot's kinematic constraints. This approach ensures that the planned path is feasible for a robot with differential drive dynamics, enabling efficient navigation in environments with obstacles. 

This project applies Dijkstra's algorithm to navigate a point robot through a defined environment with static obstacles. Dijkstra's algorithm systematically explores all possible paths from a starting point to determine the shortest path to a target location, ensuring the robot efficiently reaches its destination while avoiding obstacles. 

This project implements a complete pipeline for solving the 8-puzzle problem—a 3×3 sliding puzzle with tiles numbered 1 to 8 and a blank space (0)—using the Breadth-First Search (BFS) algorithm, and visualizes the solution as a video. It includes a solvability checker based on inversion count to verify if a given configuration can be solved, and a BFS-based solver that explores the state space level-by-level to find the shortest path to the goal configuration.

Developed an e-commerce platform on AWS with auto-scaling EC2, Application Load Balancer, RDS (Multi-AZ), and S3 integrated with CloudFront. Enhanced security with AWS WAF, Shield, and IAM, and used CloudWatch for monitoring. Managed infrastructure via Cloud Formation for scalability, performance, and cost efficiency.

Project Presentation Slide Deck : Link

Project Documentation : Link

Built a cloud-native platform using AWS EKS, Docker, and Helm for scalable deployments. Integrated Prometheus and Grafana for monitoring and alerting with AWS SNS. Automated workflows with a CI/CD pipeline using GitHub Actions, ensuring efficient deployment and rollbacks. Leveraged CloudFormation and modularized YAML for resource management.

Project Presentation Slide Deck : Link

Project Documentation : Link