Projects

Autonomous Vehicle Project

Designed a Multi-Agent System (MAS) for autonomous vehicle navigation using reinforcement learning.

Go2 Quadruped Project

Developed obstacle detection algorithms and real-time control for a quadruped robot using LiDAR and ROS2.

FaceRec Project

Developed a face recognition system with anti-spoofing mechanisms for security applications.

Bird’s Eye View Project

Researched and implemented techniques for transforming camera images into semantically segmented bird’s eye views.

VegaGPT - Internal KT Bot

Developed a chatbot for internal knowledge transfer using LangChain and Open GPT models.

AI-powered LPR System

Researched image enhancement techniques for vehicle number plate recognition accuracy in AI systems.

Autonomous Vehicle Project

Reinforcement Learning Computer Vision Multi-Agent Systems

Designed and implemented a Multi-Agent System (MAS) for autonomous vehicle simulation, integrating frameworks and architectures like FLAMEGPU2, UniAD, and DaveNet. Developed a complete autonomous driving pipeline including data collection, preprocessing, training, and evaluation using advanced computer vision models like ResNet, RegNet, and feature extraction trough BiFPN.

Built a custom Python sensor fusion module that integrates LiDAR, RADAR, Camera, GPS, and CAN bus data, streamlining perception model development and enabling real-world deployment of AI-powered driving systems.

View Code Live Demo

Go2 Quadruped Project

ROS2 LiDAR Robotics

Engineered a LiDAR-based obstacle detection algorithm to identify immediate obstacles, enabling the quadruped robot to halt autonomously. Developed a custom ROS2 package to deploy the algorithm via ROS nodes, ensuring efficient real-time performance and safety in autonomous navigation.

Contributed to localization and control algorithms using Nav2 for path planning, and integrated reinforcement learning (RL) models for dynamic obstacle avoidance.

View Code Live Demo

FaceRec Project

Computer Vision Face Recognition Security

Developed an advanced face recognition system integrated with anti-spoofing mechanisms to enhance security systems. Optimized for edge devices using Intel RealSense depth cameras for precise spoof detection, achieving a 98% detection accuracy.

The system is capable of operating on low-cost devices like Raspberry Pi, while ensuring high security and reliability in real-time face recognition applications.

View Code Live Demo

Bird’s Eye View Project

Computer Vision Segmentation Semantic Mapping

Conducted research on camera-to-bird’s-eye view transformation and implemented a UNetXST network to achieve semantic segmentation. This work provides a reliable method for generating bird’s eye view images from camera input, improving autonomous vehicle perception.

Achieved an 85% segmentation accuracy, enhancing the understanding of the environment for autonomous navigation systems.

View Code Live Demo

VegaGPT - Internal KT Bot

AI Chatbot LangChain

Developed a knowledge transfer chatbot to enhance internal communication within the AI department. Leveraged LangChain and Open GPT models to provide efficient and contextual responses for internal knowledge sharing.

The system facilitates automated, real-time knowledge exchange, significantly improving collaboration and efficiency.

View Code Live Demo

AI-powered LPR System

AI Computer Vision Vehicle Recognition

Conducted research on image enhancement techniques to improve vehicle number plate recognition accuracy. Contributed to building a robust CI/CD pipeline for streamlined deployment of updates and platform maintenance.

Developed AI models capable of recognizing number plates with high accuracy for real-time applications.

View Code Live Demo