EventsThe 13th International Electronic Conference on Sensors and Applications
Published
with-doi10.3390/engproc2026163079 (registering DOI)
This submission belongs to the session S8. Robotics, Sensors, and Industry 4.0 of the event The 13th International Electronic Conference on Sensors and Applications
Published date
16 Nov, 2026
Academic Editor
author-avatarFrancisco Falcone
Citation
Suman Kunwar, Timotius Alvaro Hartanto, FLAP‑SIX: A Fixed‑Flapper, Six‑Compartment Smart Bin with a Lightweight AI Model, in Proceedings of The 13th International Electronic Conference on Sensors and Applications, 18 November–20 November 2026, MDPI: Basel, Switzerland, doi: 10.3390/engproc2026163079
Share
Email
Facebook
Twitter
LinkedIn

FLAP‑SIX: A Fixed‑Flapper, Six‑Compartment Smart Bin with a Lightweight AI Model

Timotius Alvaro Hartanto 2
1. Applied AI Lab, DWaste, Baltimore, MD 21218, USA
2. Innovation Hub, Green School Bali, Badung, 80352, Indonesia
Abstract

This study proposes FLAP-SIX, a low-cost smart waste bin with six compartments that uses a flapper-based automated sorting mechanism to sort household waste. Items placed on the flapper are detected by a sensor, triggering the camera to capture an image, run inference, and redirect them into the appropriate compartment. It investigates effect of camera positioning (centered, offset) and lighting condition (natural, indoor) on waste classification, evaluates accuracy and inference speed using lightweight (MobileNetV3-S, YOLOv26N) and larger architectures (ResNet50V2, ViT-B/16), and also examines the reliability of the flapper mechanism. Models were benchmarked using precision, recall, F1-score, inference latency, model size, and throughput, and the best-performing model was deployed on the constructed FLAP-SIX bin. Experiment results showed that centered camera positioning significantly improved classification performance, reaching up to 62% accuracy compared to inclined position, which reduced accuracy by 21% under indoor bulb lighting. Indoor lighting conditions performance was more stable with consistent results than natural lighting under deployment.

Benchmarking revealed that lightweight models achieved better real-time performance and lower model sizes (5.38 - 6.24 MB), with MobileNetV3-S achieving 95.56% accuracy at 6.89 ms latency and 145.22 FPS. Larger architectures (ResNet50V2, ViT-B/16) showed lower classification performance even with larger model sizes (273.14 - 343.28 MB), with ResNet50V2 achieving 70.83% accuracy at 10.65 ms latency, highlighting trade-off between model complexity, deployment efficiency and inference performance. Under centered-camera position, the flapper mechanism achieved a success rate of 53% but performed poorly under an inclined camera at natural lighting. The stationary placement design reduced motion blur and improved image consistency, and the flapping mechanism eliminates the need for multiple bins or the need of high-throughput carousel, demonstrating the potential of FLAP-SIX for household waste sorting.

Keywords
smart waste bin
automated waste sorting
edge AI
real-time classification
flapper mechanism
household recycling
Manuscript
Sciforum-191751engproc-163-00079.pdf
Design and simulation evaluation of an embedded multi-sensor ADAS architecture with AI-assisted collision warning
Precision Agriculture Through Machine Learning and IoT: A Novel Framework for Efficient Irrigation Management