AI-Powered Marine Cleanup · Autonomous Robotics

Swamn

An intelligent autonomous robotic system that detects and collects floating marine debris — preserving our oceans, one wave at a time.

View Results ↓ How It Works
14M Tonnes of Plastic Enter Oceans / Year
1M+ Seabirds Killed by Plastic Annually
700+ Marine Species Affected
2040 Triple Ocean Plastic Without Action
🌊
The Crisis

Oceans drowning in plastic

  • 14M Metric tonnes of plastic enter our oceans every single year, threatening marine life and human food security across the globe.
  • 1.8T Plastic pieces float in the Great Pacific Garbage Patch alone — 100,000 metric tonnes — with dozens of such patches worldwide.
  • 700+ Marine species including fish, seabirds, dolphins and turtles are directly harmed by plastic pollution every year.
  • 2040 Without urgent intervention, annual ocean plastic input could reach 23–37 million metric tonnes per year, a near-tripling of today's crisis.
The Scale of the Problem

The ocean can't wait

Every minute, the equivalent of a garbage truck of plastic is dumped into our oceans. Here's the true cost.

🐠 Marine Life 700+
Marine species are directly affected by plastic — from sea turtles ingesting bags to dolphins entangled in nets.
🐦 Seabirds 1M+
Seabirds die each year after ingesting plastic fragments, mistaking them for food. Chicks are fed plastic pieces by unknowing parents.
🐢 Sea Turtles 100%
All 7 sea turtle species are affected by plastic pollution — entanglement and ingestion are leading causes of their decline.
🌊 Ocean Surface 5.25T
Plastic pieces estimated floating in our oceans today. Surface plastics are the most critical — reachable by robots like Swamn.
🏖️ Coastlines 620K km
Of global coastline are contaminated with plastic debris — impacting coastal communities, tourism, and shoreline ecosystems.
🔬 Microplastics 24T
Microplastic particles now found in the deep ocean. They enter the food chain — from plankton to fish, to our dinner plates.
System Architecture

Three layers of intelligence

Swamn operates as a layered autonomous system, combining edge computing, AI inference, and motor control for continuous marine cleanup.

👁️
Perception Layer
Pi Camera Module v2 streams live frames to a ResNet18 + KNN inference pipeline. Each frame is classified as GARBAGE or CLEAR every ~500ms, triggering immediate motor decisions.
ModelResNet18 + KNN
Camera8MP · 1080p@30fps
Inference Rate~2 Hz
🧠
Control Layer
Raspberry Pi 5 acts as the central brain, processing AI decisions and sending commands to motor controllers and the belt-net collection mechanism in real time.
ComputeRPi 5 · 4GB RAM
Motor MCUESP32 WROOM-32
GPS MCUESP8266 @ 1Hz
⚡
Power Layer
A 12V SLA battery powers all systems. When voltage drops below 11.5V, Swamn autonomously returns to its solar docking station to recharge — fully off-grid.
Battery12V 7Ah SLA
Solar Panel10W Monocrystalline
Runtime5.5 Hours
AI Pipeline

From pixel to action

📸
Capture
Pi Camera streams 1080p frames at 30fps to Raspberry Pi 5
🔄
Preprocess
Frames resized, cropped & normalized for inference
🤖
Classify
ResNet18 extracts features; KNN classifies as GARBAGE or CLEAR
🧭
Navigate
Pi 5 sends motor commands — approach, scan, or return to dock
🪣
Collect
Belt-net conveyor scoops debris in ~11.4 seconds average
The Big Upgrade

Swamn Intelligence

Multiple bots. One mission. A coordinated fleet of Swamn units that communicate, divide area, and clean at a scale no single robot can match.

🤖 🤖 🤖 📡 BOT-01 LEAD BOT-02 BOT-03 HUB
01
Mesh Communication
Each Swamn bot broadcasts its GPS position, debris density data, and battery level to all others via a Wi-Fi mesh network — every bot knows what every other bot is doing.
02
Automatic Area Division
The fleet autonomously partitions the cleanup zone into sectors. As bots finish their sector, they redistribute across uncleaned areas — maximising coverage with zero overlap.
03
Collective Intelligence
Bots share detection data in real time. A debris sighting by one bot immediately alerts the nearest bot to respond — creating a living, adaptive cleanup network.
04
Lead Bot Coordination
A designated lead bot runs the scheduling algorithm, assigning tasks based on each unit's remaining capacity and proximity — like a fleet manager on the water.
05
Fault Tolerance
If a bot returns to dock or goes offline, remaining bots automatically rebalance the workload — the mission continues even when individual units are unavailable.
06
Scalable Fleet
Start with 2 bots, scale to 200. The swarm protocol requires no changes — each additional bot improves collective coverage linearly, making fleet expansion effortless.
On-board Processing

True autonomy at the edge

The next Swamn ditches external dependency entirely. All intelligence runs on-board — real-time, resilient, and ready for the real world.

🍓
Raspberry Pi 5
Quad-core ARM Cortex-A76 at 2.4GHz — powerful enough to run optimized YOLO models locally at real-time speeds.
🪸
Google Coral TPU
Dedicated ML accelerator co-processor capable of 4 TOPS — runs YOLO inference at 30+ FPS with minimal power draw.
🧠
Intel Neural Compute Stick 2
USB-attached neural accelerator with 4GB LPDDR4. Supports INT8 inference with OpenVINO-optimized YOLO models at scale.
☁️
Cloud Only When Needed
Edge AI decides — only unusual detections, alerts, and daily summaries are sent to cloud. 95% of work stays on the bot.

Why Edge AI Changes Everything

Current prototypes depend on Wi-Fi and external servers — fine for testing, but fragile in the real world. Rivers, coastal areas, and remote lakes have no reliable connectivity.

With edge AI, Swamn makes detection, navigation, and collection decisions entirely on-board. It can clean for hours in isolated locations without any connection whatsoever.

YOLO Local No Wi-Fi Needed INT8 Inference OpenVINO 4 TOPS Real-time
Performance

Prototype results

91.7%
Overall Classification Accuracy
False Positive Rate
5.0%
False Negative Rate
9.0%
11.4s
Mean Collection Time
Range: 8.2 – 15.7s
3.8m
Full Area Coverage
360° scan of 1.5×1.0m tank
92%
Docking Success
23/25 first-attempt trials
5.5h
Continuous Operation
Solar-assisted battery life
Debris Category Trials Accuracy
PET Bottles20
100%
Polystyrene Foam20
95%
Bottle Caps20
95%
Plastic Bags20
85%
Drinking Straws20
80%
Build

Engineering the machine

Every component chosen for cost-effectiveness, reliability, and marine-grade durability. Built for ₹17,500 — a fraction of commercial alternatives.

🖥️ Processing & Comms
Raspberry Pi 54GB · quad-core 2.4GHz
ESP32 WROOM-32240MHz · Wi-Fi · ×2
ESP8266 NodeMCUGPS & telemetry MCU
📷 Sensing
Pi Camera Module v28MP · 1080p@30fps
GY-GPS6MV2u-blox NEO-6M · UART
MPU6050 Gyroscope6-DOF IMU · I2C
⚙️ Actuation & Drive
DC Motors ×212V · 150RPM · waterproof
Dual BTS796043A peak H-bridge
Belt-Net SystemCustom rubber conveyor
🔋 Power System
SLA Battery12V 7Ah sealed lead-acid
Solar Panel10W monocrystalline
PWM Charge Controller12V / 10A
🚢 Structure & Docking
Chassis2mm mild steel · epoxy paint
Docking ContactsMagnetic pogo-pin · gold-plated
Debris Bins2L onboard + 20L dock
120 Test Trials
$210
Total prototype cost
(vs $5K–$50K commercial)
Scale

Real-world impact at scale

A small fleet of low-cost Swamn units can recover hundreds of tonnes of surface macroplastics annually when focused on pollution hotspots.

Fleet Impact Calculator — 100 Units
Units deployed100
Collection per unit/day5 kg
100 × 5500 kg/day
500 × 365182,500 kg/yr
Annual Impact 182.5 T/yr
Aligned with Global Goals
14
Life Below Water Reducing plastic pollution in marine ecosystems to protect biodiversity
12
Responsible Consumption Enabling circular waste recovery and responsible plastic management
7
Clean Energy Powered entirely by solar energy for zero-emission, off-grid operation
Innovators

The minds behind Swamn

RS
Rishi Singh
Lead Developer & Creator

16-year-old innovator passionate about using technology to solve real-world environmental challenges. Creator and lead developer of Swamn, integrating ML models with Raspberry Pi and ESP32 hardware for real-time garbage detection.

Selected among 1.5 lakh students at IIT Guwahati
Recognised at district level — INSPIRE Awards
Registered for IIT Delhi competition
VR
Vaibhav Raj
Co-developer & ATL VP

16-year-old Class XI student with a strong inclination toward academics, leadership, and technology. Vice-President of the ATL (Atal Tinkering Lab) Club, actively organising events, mentoring peers, and fostering creativity.

Vice-President, ATL Club — Sunbeam School
Olympiad achiever in multiple subjects
Strong analytical & problem-solving skills
Sunbeam School Mughalsarai
Uttar Pradesh, India · Class XI · 2026
INSPIRE Award
IIT Guwahati
IIT Delhi
ATL Club
Roadmap

What comes next

The current prototype is just the beginning. Future iterations of Swamn will expand capability, autonomy, and scale.

01
Submerged Debris Detection
Expanding the sensing stack with sonar and underwater cameras to detect and recover plastics below the surface, targeting microplastic accumulation zones.
02
Turbulent Open-Water Testing
Moving beyond controlled tank conditions to real coastal environments — harbors, rivers, and estuaries — with wave-compensation and ruggedized hull.
03
Full Swarm Deployment
Deploying coordinated fleets of Swamn units with the swarm intelligence protocol, sharing detection data and optimizing coverage across large water bodies in real time.