Explore My Projects and Skills

Projects

National Finalist

UAV YOLOv8 RaspberryPi

NIDAR 2026 – Dual-Drone Rescue System

National finalist in MeitY + DFI's disaster management challenge. Autonomous scout + delivery drones with YOLOv8 detection (91% accuracy) and custom Python GCS.

NIDAR 2026 - Autonomous Multi-Drone Disaster Response System

National Finalist

Competition Overview

NIDAR (National Innovation in Drone Applications & Research) is India's largest national drone innovation challenge hosted by the Ministry of Electronics and Information Technology (MeitY) and Drone Federation India (DFI) under the SwaYaan Initiative. The finals took place from January 10-16, 2026 at Gautam Buddha University, Greater Noida.

🏆 Prize Pool: ₹40 Lakhs + Startup Incubation + Industry Mentorship
🎯 Mission: Disaster Management - Flood Rescue Operations
👥 Team: Swatantra System (N250771)
🏫 Institute: Shah & Anchor Kutchhi Engineering College
📍 Coverage Area: 1 km² (30 Hectares) Autonomous Scanning

Problem Statement

A coastal town is flooded after 48 hours of continuous rainfall. Residents are trapped on rooftops without food, water, or medical supplies. Traditional rescue methods fail due to collapsed roads, fast-moving water, and communication blackouts. Every minute is critical in the "Golden Hour" of survival.

Mission Objectives:

🔍 Scan 30 hectares systematically using lawn-mower grid pattern
🧠 ML-based survivor detection with YOLOv8 computer vision (60-70% field accuracy)
📍 GPS geotagging with 2.5m precision using Holybro M8N module
📦 Autonomous payload delivery to 10 survivors per mission
🤖 Single-operator control with KML-based mission planning

Solution Architecture - Distributed Autonomous System

🔍 Scout Drone (Quadcopter) - Intelligence & Reconnaissance

Primary Function: Area scanning with autonomous grid coverage pattern at 30-50m altitude

Detection System: YOLOv8 real-time inference on Raspberry Pi 5 (238ms per frame, 91% lab accuracy)

Output: Geotagged survivor queue with coordinate transformation and multi-frame validation

🚁 Delivery Drone (Hexacopter) - Payload Deployment

Primary Function: Queue-based autonomous navigation to survivor coordinates

Mechanism: 5-hook sequential release system with MG995 servo actuation (200g per payload)

Mission End: Queue exhaustion → Automatic Return to Launch (RTL)

💻 Ground Control Station (GCS)

Interface: Python-based custom GCS with real-time map visualization

Mission Planning: KML file upload for polygonal area definition

Monitoring: Live telemetry, queue status, emergency abort capability

Hardware Specifications

Component	Scout Drone (Quadcopter)	Delivery Drone (Hexacopter)
Flight Controller	Pixhawk 2.4.8	Pixhawk 2.4.8
Motors	Emax 935 KV Brushless	Flycat 5010 iRotor 360 KV
ESC	ReadyToSky 30A	Hobbywing 40A
Propellers	10×4.5 inch	14-inch
Battery	14.8V 9000mAh 4S Li-ion	14.8V 12600mAh 4S Li-ion
Total Weight	~2.15 kg	~2.5 kg (base)
Flight Time	~27 minutes	~10 minutes (with payload)
GPS Module	Holybro M8N (2.5m accuracy)	Holybro M8N (2.5m accuracy)
Companion Computer	Raspberry Pi 5	Raspberry Pi 3
Telemetry	433MHz SiK V3 (100mW)	433MHz SiK V3 (100mW)
Camera	Raspberry Pi Camera Module	-
FPV System	600 TVL + 5.8GHz TX	-
Payload Mechanism	-	MG995 Servo + 5-Hook System

YOLOv8 Computer Vision Pipeline

Detection System

Model: YOLOv8 optimized for Raspberry Pi 5 hardware acceleration

Resolution: 900×900 pixels for optimal accuracy-speed balance

Inference Speed: 238ms per frame

Lab Accuracy: 91% (Precision: 84%, Recall: 98%, mAP@0.5: 0.97)

Field Accuracy: 60-70% in real flood conditions

Robustness Mechanisms

✅ Confidence Filtering: Threshold >0.7 to eliminate false positives
✅ Multi-Frame Validation: 3+ consistent detections required
✅ Lighting Variation Handling: Temporal consistency mitigates shadows
✅ Water Reflection Filtering: Multi-frame validation filters glare

Coordinate Transformation Pipeline

1️⃣ Image detection (YOLOv8 bounding box) → 2️⃣ Georeference (GPS, altitude, camera orientation) → 3️⃣ Ground plane calculation (ray casting) → 4️⃣ GPS coordinate output (Lat/Lon with confidence)

Total Positioning Error: ±3.0m RMS (GPS: ±3.5m, Image: ±1.0m, Altitude: ±0.5m)

Operational Workflow

Mission Timeline (30 minutes total)

Pre-Flight Setup: 5 minutes (GPS lock, calibration, mission upload)

Scout Coverage: ~25 minutes (1 km² at 0.04 km²/min, 30-50m altitude)

Delivery Mission: ~15 minutes (5 targets with sequential deployment)

RTL & Recovery: 5 minutes

Flight Parameters

Scout Pattern: Lawn-mower grid coverage (20-30m track spacing, 5-8 m/s speed)

Delivery Approach: 15-20m altitude, 5-10m loiter, 1-2 m/s descent, 5-10m release altitude

Safety: 25% battery reserve, geofencing (500m radius, 100m ceiling), automatic RTL on low battery

Key Performance Metrics

🎯 Detection Accuracy: 60-70% in field conditions (91% in controlled lab)
📍 GPS Positioning: 2.5m accuracy (Holybro M8N, 3-GNSS support)
⏱️ Mission Duration: 30 minutes end-to-end
📦 Payload Capacity: 10 targets per mission (5 items per delivery drone sortie)
✅ Release Reliability: 100% (50+ bench test cycles)
🔋 Flight Time: Scout 27min, Delivery 10min (with payload)
📡 Communication Range: 433MHz telemetry (100mW, >1km range)

Safety & Fail-Safe Systems

Battery Monitoring

Auto RTL at 25% Critical RTL at 15% Per-Cell Voltage Monitor

Geofencing

500m Radius Hard Limit 100m Altitude Ceiling Breach → Immediate RTL

Payload Safety

Instability Lock (±2g) RTL Retention Manual Abort Safety

Communication Loss

Heartbeat Monitor >3s Loss → RTL Queue-Based Recovery

Team Contributions

👨‍💼 Parth Karanjawane (Leader): Components Selection & System Architecture
🔧 Prathamesh Aarya: Hardware Assembly & Integration
🔧 Manav Kothari: Hardware Design & Testing
🧪 Shivam Yadav: Software Development, Assembly, Testing & System Integration
🧪 Yashwant Mummidi: Assembly, Testing & Field Validation
💻 Tanish Nagarkar: Software Development (GCS & Mission Planning)
💻 Pratyush Mishra: Software Development (YOLOv8 Pipeline)

Challenges & Learnings

The system was fully operational and extensively tested through 20+ software-in-loop simulations with 95% success rate. During the national finals, we successfully demonstrated dual-drone coordination, YOLOv8 detection, and autonomous navigation. Minor technical challenges provided invaluable insights into:

🔧 Real-world autonomous drone operations at national competition scale
🤝 Multi-drone coordination with queue-based synchronization
🧪 Field testing under pressure with 30-minute mission constraints
💡 National-level exposure with MeitY officials and industry mentors
📚 Advanced MAVLink integration and custom GCS development
🎯 YOLOv8 optimization for edge computing (Raspberry Pi 5)

Project Resources

View Certificate NIDAR Official Website

1 / 12

Featured

Flask MongoDB AI/ML Python

NutriAI – AI Nutrition Advisor

AI-powered nutrition platform with JWT auth, 650K+ foods, ML insights, and real-time analytics.

Live Demo Code

NutriAI - AI-Powered Nutrition Advisor

Live on Render

Project Overview

NutriAI is a full-stack nutrition tracking platform powered by advanced machine learning algorithms. Users can log meals, track macros (protein, carbs, fats), set personalized goals, and receive AI-powered dietary insights. The platform analyzes eating patterns using diet classification, K-Means clustering, and LSTM forecasting to provide comprehensive nutrition recommendations.

Core Features

Secure Authentication

JWT-based authentication with bcrypt password encryption for maximum security

650K+ Food Database

Comprehensive nutrition data integrated via FatSecret API

Real-time Dashboard

Interactive visualizations with Chart.js for tracking progress

Goal Tracking

Set and monitor daily calorie and macro targets with progress bars

Streak Monitoring

Track consecutive logging days for motivation and consistency

Historical Analysis

View daily, weekly, and monthly nutrition trends and patterns

Machine Learning Models

1. Diet Classification Algorithm

Type: Rule-based classification using macro ratios

Accuracy: 85-95% classification accuracy

🥩 High Protein: >35% protein intake
🥑 Ketogenic: >60% fat, <10% carbs
🥗 Low Carb: <25% carbohydrate intake
⚖️ Balanced Diet: 15-25% protein, 45-65% carbs

2. K-Means Clustering Analysis

Algorithm: K-Means clustering with 5 distinct user profiles

Features Analyzed: Daily calories, macro distribution, meal timing patterns, food diversity

💪 Health-Conscious Eaters: Balanced nutrition focus
🏋️ High-Protein Athletes: Performance-driven diet
🍝 Carb-Lovers: High carbohydrate preference
⚖️ Balanced Enthusiasts: Moderate macro distribution
🔄 Flexible Dieters: Variable eating patterns

3. LSTM Forecasting Model

Model Type: Long Short-Term Memory (LSTM) Neural Network

Purpose: Predict future calorie intake based on historical eating patterns

Data Requirements: Minimum 14 days of consistent logging for accurate predictions

Applications: Meal planning, calorie budget recommendations, trend identification

Technology Stack

Backend

Flask 3.0 Python 3.11 MongoDB Atlas JWT Auth Gunicorn bcrypt

Machine Learning

scikit-learn NumPy pandas LSTM K-Means

Frontend

HTML5 CSS3 JavaScript ES6+ Chart.js Responsive Design

APIs & Deployment

FatSecret API Render Cloud RESTful API MongoDB Atlas

System Architecture

Backend API: Flask-based RESTful API handling authentication, meal logging, and analytics

Database: MongoDB Atlas for scalable NoSQL data storage

ML Pipeline: Scikit-learn models for diet classification and user clustering

Frontend: Responsive single-page application with real-time chart updates

Deployment: Render cloud platform with Gunicorn WSGI server

Project Metrics

📊 Database Size: 650,000+ food items with complete nutrition data
🎯 ML Accuracy: 85-95% diet classification accuracy
⚡ API Response: Average 200ms response time
🔐 Security: JWT tokens with 24-hour expiration
📱 Responsive: Fully optimized for mobile, tablet, and desktop

Future Enhancements

🍽️ Meal recommendation system based on dietary preferences
📸 Image-based food recognition using computer vision
👥 Social features for community support and challenges
🔔 Push notifications for meal reminders and goal milestones
📈 Advanced analytics with predictive health insights

Project Links

Live Demo GitHub Repository

Autonomous Drone – ISRO Robotics Challenge 2024

Developed an autonomous drone for GPS-denied environments using Pixhawk & Raspberry Pi. Selected in qualification round by ISRO.

Robotic Arm – Web Controlled using ESP32

Built a robotic arm with 4-DOF and recording-playback via a web interface. Secured 2nd Prize in college-level robotics competition.

View Certificate

Multiple Frontend Projects

Created and deployed frontend websites using HTML, CSS, JavaScript, React.js. Includes portfolios and product pages.

Digital Thermometer – Mini Project

Arduino-based mini project for real-time temperature monitoring with LCD display.

Hi, I'm Shivam Yadav, a Third-Year B.Tech ECE Student.

Projects

NIDAR 2026 – Dual-Drone Rescue System

NutriAI – AI Nutrition Advisor

Autonomous Drone – ISRO Robotics Challenge 2024

Robotic Arm – Web Controlled using ESP32

Multiple Frontend Projects

Digital Thermometer – Mini Project

Achievement Gallery

Gold Medal – Academic Excellence

NIDAR 2026 - National Finalist

Felicitated by ISRO Scientists

ISRO Robotics Challenge 2024

2nd Rank – Robotic Arm Project

Professional Certifications

My Skillset

HTML / CSS

JavaScript

React.js

MongoDB

Python

Let's Connect