Archived Projects
Complete technical portfolio showcasing research and development across computer vision, robotics, and machine learning
This archive contains my complete collection of technical projects, spanning from cutting-edge research implementations to production-ready systems. Each project represents a deep exploration of methodologies at the intersection of artificial intelligence, robotics, and computer vision. From real-time SLAM algorithms to foundation model integrations, these projects demonstrate both theoretical understanding and practical engineering capabilities.
Dense Stereo Reconstruction in Real-time
Implementation of epipolar geometry and block-matching for disparity map computation using KITTI dataset. Advanced computer vision techniques for stereo rectification with ORB-SLAM 3 pose estimation.
More details- Real-time 3D reconstruction for robotics applications
- KITTI dataset integration with performance optimization
- ORB-SLAM 3 pose estimation implementation
- Advanced stereo rectification algorithms
Neural Machine Translation
Comparative analysis of LSTM and GRU-based Bahdanau Attention architectures for sequence-to-sequence translation. English to Hindi translation system with ADAM optimization.
More details- BLEU score metrics for architectural effectiveness analysis
- Bahdanau Attention mechanism implementation
- LSTM vs GRU comparative performance study
- ADAM optimization for enhanced convergence
Pose-Graph Optimization
Backend optimization framework for SLAM with graph-based robot trajectory modeling. Non-linear least squares optimization using Jacobian matrices and weighted loop-closures.
More details- Robust localization and mapping capabilities
- Graph-based robot trajectory modeling
- Jacobian matrices for optimization
- Weighted loop-closure detection and correction
Artpark Robotics Challenge
Second place in nationwide robotics competition with computer vision, navigation, and control integration. Real-time trash detection using RGB-D cameras with ROS navigation stack.
More details- Autonomous environmental interaction and task execution
- RGB-D camera integration for depth perception
- ROS navigation stack implementation
- Real-time computer vision processing
Augmented-Reality and Robot Assisted Knee Surgery
Comprehensive system development for AR-guided robotic knee surgery. Gathered requirements from user studies, market analysis, and medical professionals to inform system architecture and implementation.
More details- Apple Vision Pro integration for 3D spatial computing
- ICP-based bone model detection and tracking
- Real-time 3D and RGB information processing
- Medical robotics system integration
3D Foundation-Models for Monocular Video Reconstruction
Advanced 3D reconstruction system using foundation model embeddings with semantic-geometric feature fusion. Implemented cross-attention mechanisms for enhanced camera extrinsics recovery.
More details- DINOv2 and Depth Anything foundation model integration
- Hierarchical state representation for temporal consistency
- DUST3R-style architecture for robust 3D reconstruction
- Adaptive keyframe selection with confidence-aware refinement
Neural-Assisted Depth Disparity Estimation
Developed coarse-to-fine network architecture for OAK-D Pro stereo camera system. Achieved top 25 ranking internationally in Luxonis Hackathon competition.
More details- Real-world disparity estimation quality optimization
- Onboard compute and FPS constraint optimization
- Coarse-to-fine neural network architecture
- International hackathon competition success
Music, Mental Health, and Representation Learning
Applied BERT-based sentiment analysis and k-means clustering to uncover nuanced relationships between language patterns and acoustic music features during COVID-19 pandemic.
More details- Mental health subreddit data mining and analysis
- BERT-based sentiment analysis implementation
- K-means clustering for pattern discovery
- Acoustic feature extraction and correlation analysis
Autonomous Robot Navigation System
Complete autonomous navigation pipeline with obstacle avoidance, path planning, and SLAM integration. Deployed on real robots with sensor fusion and real-time decision making.
More details- Multi-sensor fusion for robust localization
- Dynamic obstacle avoidance algorithms
- A* and RRT* path planning implementation
- Real-time control system integration
Multi-Agent Coordination and Swarm Robotics
Distributed coordination algorithms for robot swarms with consensus protocols and formation control. Implemented Byzantine fault tolerance and emergent behavior patterns.
More details- Distributed consensus algorithms implementation
- Formation control and flocking behaviors
- Byzantine fault tolerance in multi-agent systems
- Emergent behavior simulation and analysis