Project 2: Swarm Robotic System for Garbage Detection,
Pick-and-Place & Disposal
A swarm of low-cost mobile rovers patrols a defined area (campus, park, street section), using an onboard
camera and lightweight AI model to detect litter and classify it as disposable (wet/biodegradable) or
non-disposable (dry/recyclable) waste. The nearest free rover approaches the item, picks it up with a small
robotic arm and gripper, and transports it to the nearest matching dustbin. Rovers coordinate over a wireless
mesh link so that only one rover claims a detected item, avoiding duplicate pickups.
Chapter Summaries
Chapter Content Summary
Ch 1 – Introduction Problem definition: manual litter collection is slow and labour-intensive. Concept of a
multi-rover (swarm) system that detects, classifies and disposes of garbage
autonomously.
Ch 2 – Architecture Rover hardware layout (chassis, sensors, arm), onboard waste-classification model
(lightweight CNN / YOLOv8-nano), wireless swarm communication (ESP-NOW /
NRF24L01 mesh), and the nearest-rover task-dispatch algorithm.
Ch 3 – Detection & Camera-based item detection and bounding-box localisation, wet-vs-dry /
Classification recyclable-vs-non-recyclable classification, ultrasonic/IR distance estimation for arm
approach, and moisture/metal sensor fusion to improve classification confidence.
Ch 4 – Pick-and-Place 4-DOF servo arm kinematics and gripper control, grasp confirmation, navigation to the
& Disposal nearest matching dustbin, drop-off confirmation sensor, and swarm collision/task-conflict
avoidance.
CEP Analysis
P Aspect Applies Justification
Code ?
P1 * Depth of Knowledge YES Computer-vision/CNN classification theory, servo-arm inverse
kinematics, and swarm task-allocation algorithms applied together in
one embedded system.
P2 Conflicting YES Battery life vs. continuous patrol/scanning; arm payload capacity vs.
Requirements rover size/cost; classification accuracy vs. on-board processing
power of a low-cost MCU.
P3 Depth of Analysis YES No closed-form solution for jointly optimising patrol coverage,
classification threshold, arm-approach path and nearest-dustbin
routing across the swarm.
P4 Novel / Infrequent YES Waste-type classification for mixed local litter (wet/dry,
Issues plastic/metal/paper) needs a custom-labelled dataset; reliable
pick-and-place of irregular litter is non-standard.
P5 Beyond Standards PARTIA No IEC/ISO standard governs multi-robot litter collection; the swarm
L communication and task-dispatch protocol is custom-designed
beyond textbook scope.
P6 Stakeholder Diversity YES City corporation / facility management (cleanliness, cost), citizens
(clean public space), environmental regulators (waste segregation),
maintenance staff (bin servicing).
P Aspect Applies Justification
Code ?
P7 Interdependence of YES Detection accuracy drives arm-targeting success; mesh bandwidth
Sub-problems affects task-dispatch latency; battery/payload trade-offs limit patrol
range and number of pickups per charge.
CEP Verdict: P1 (Mandatory) + P2 + P3 + P4 + P5 (Partial) + P6 + P7 -> CEP confirmed.
Sensors & Components Required (per Rover)
All parts below are commonly stocked by Bangladeshi robotics/electronics retailers such as RoboticsBD ([Link]),
Robotech Shop BD, City Tech BD (physical shop at Ibrahim Electric & Electronics Market, Nawabpur, Dhaka), Udvabony,
TechShopBD, BOT Engineers BD and [Link], or in person at the Stadium Market / Elephant Road / Nawabpur
electronics markets in Dhaka — making this build fully feasible as a local student demo project.
Function Component / Sensor Purpose in this Project BD Sourcing Notes
Vision & ESP32-CAM module (or Captures images of litter; runs a Widely stocked; ESP32-CAM
Classification Raspberry Pi 4 + Pi lightweight CNN / YOLOv8-nano is the budget option, Pi 4 +
Camera) (TensorFlow Lite) model to detect the camera gives more inference
item and classify it as disposable headroom.
(wet) or non-disposable
(dry/recyclable).
Distance / HC-SR04 Ultrasonic Obstacle avoidance during patrol and One of the most common BD
Obstacle Sensor (x2–3) distance measurement while stock items, very low cost.
approaching a detected item.
Close-range IR Proximity / Confirms the item is within gripper Common IR obstacle
Pickup Object-presence Sensor range immediately before the arm modules; widely available.
(x1–2) closes on it.
Material Cue Inductive Proximity / Metal Flags metal litter (cans, foil) as Available as an
Detector Module non-disposable/recyclable, improving automation-grade proximity
classification reliability beyond sensor.
camera vision alone.
Material Cue Capacitive Moisture Auxiliary wet-vs-dry cue (e.g. food Sold as soil-moisture sensor
Sensor waste vs. paper/plastic) to support modules; works the same
the camera classifier. way for this purpose.
Manipulation 4-DOF Robotic Arm Kit Picks up the detected litter item and Ready-made acrylic/wood
(SG90 / MG996R servos) places it into the rover's carrying bin DIY arm kits are sold locally
+ 2-finger gripper or directly into the dustbin. with all servos included.
Manipulation PCA9685 16-Channel Drives all arm servos smoothly and Standard module, easily
PWM Servo Driver accurately from the main controller available.
over I2C.
Locomotion 4WD/2WD Chassis + Provides the rover's mobility for Pre-cut acrylic chassis kits
N20/TT DC Gear Motors patrolling and travelling to the item (including locally made ones)
(x4) + Motor Driver and the dustbin. are common BD stock.
(L298N / TB6612)
Navigation 8-Channel IR Guides the rover along a fixed demo Both are standard catalogue
Line-Tracking Array route to known dustbin points; GPS items; line array is simplest
(indoor demo) or NEO-6M variant allows real outdoor for a classroom demo.
GPS (outdoor / advanced) geo-tagging of dustbin locations for
the swarm.
Swarm Comm NRF24L01+ Module (or Lets rovers broadcast 'item claimed' / NRF24L01+ is cheap and
unication ESP-NOW on ESP32) 'task complete' messages so only common; ESP-NOW needs
one rover handles each detected no extra module if ESP32 is
item, and so idle rovers know which used.
areas are already covered.
Drop-off IR Break-beam Sensor or Confirms the item actually landed in Same HC-SR04 / IR modules
Confirmation Mini Ultrasonic Sensor the bin and can report approximate as above can be reused for
(mounted in dustbin) bin fill-level. this.
Power 11.1V 3S Li-ion/LiPo Powers motors, arm servos and LiPo packs, BMS boards and
Battery Pack + BMS + electronics; buck converter steps buck converters are stocked
LM2596 Buck Converter voltage down for the logic boards. by all listed BD retailers.
Function Component / Sensor Purpose in this Project BD Sourcing Notes
Main Arduino Mega 2560 Runs sensor fusion, classification Both options are common;
Controller (motors + arm I/O) paired trigger, arm sequencing and Mega is cheaper, Pi 4 is more
with ESP32-CAM (vision), swarm-communication logic. capable.
or a single Raspberry Pi 4
for both
Optional TCS3200 Colour Sensor Adds colour-based cues to separate, Inexpensive add-on, available
Refinement e.g., clear/coloured plastics for at the same stores.
finer-grained sorting demos.
Status Buzzer + RGB LED Indicates rover state (searching / Trivial-cost items, available
Feedback approaching / picking / delivering) for everywhere.
the live demo audience.
Indicative Demo-Scale Bill of Materials (per Rover)
Approximate Bangladesh retail prices for a single demo rover, budget-friendly options chosen (prices vary by vendor/availability and
are meant only as planning estimates).
Item Qty Approx. Unit Price Approx. Subtotal
(BDT) (BDT)
Arduino Mega 2560 / ESP32 Dev Board (controller) 1 850 – 1,400 1,400
ESP32-CAM module 1 900 – 1,200 1,200
HC-SR04 Ultrasonic Sensor 3 100 – 150 450
IR Proximity Sensor Module 2 60 – 100 200
Inductive Proximity / Metal Sensor 1 250 – 400 400
Capacitive Moisture Sensor 1 120 – 180 180
4-DOF Robotic Arm Kit + Gripper (4x servo) 1 set 1,800 – 2,500 2,500
PCA9685 16-Ch Servo Driver 1 350 – 500 500
4WD Chassis Kit + N20/TT Motors (x4) 1 set 1,500 – 2,200 2,200
L298N / TB6612 Motor Driver 1 150 – 300 300
8-Channel IR Line Sensor Array 1 350 – 500 500
NRF24L01+ Wireless Module 1 150 – 250 250
11.1V LiPo Battery + BMS 1 1,200 – 1,800 1,800
LM2596 Buck Converter, wiring, chassis hardware, — — 700
misc.
Estimated total per rover ≈ 12,500 – 13,500
For the swarm demo, 2–3 identical rovers are sufficient to demonstrate task-claiming and collision-avoidance behaviour without a
large budget increase, since most of the cost is per-rover hardware that simply repeats.