Towards an Automated Waste-Sorting System: Computer Vision and Neural-Network Inverse Kinematics for a 5-DOF Arm

Automated waste sorting with a robotic manipulator requires two capabilities that belong to different fields: recognizing and localizing the object in the image, and converting a target Cartesian position into joint angles. This paper reports the construction and separate evaluation of both modules for a 5-DOF manipulator, and verifies the inverse-kinematics module on a continuous trajectory. For the recognition module a dedicated dataset, EPU_waste, of 2,955 images and 4,068 annotated instances in three classes was built, and two instance-segmentation models, YOLOv8s-seg and YOLO11s-seg, were trained and compared on it. YOLOv8s-seg reaches a recall of 0.97 against 0.88 and runs at 56.5 frames per second on a Quadro RTX 3000 GPU, and was therefore selected. For the inverse-kinematics module a 7–15–15–4 feedforward network with 424 weights was trained by the Levenberg–Marquardt algorithm over ten independent runs, giving a joint angle error of 2.2164 ± 0.1811° and a position error of 9.96 mm. On a continuous circular trajectory, the network gives a position RMS error of 3.461 mm with a standard deviation of only 1.30 mm, so the error is largely a near-constant systematic offset.

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-19
DOI
https://doi.org/10.5281/zenodo.24359445
Primary Topic
Advanced Neural Network Applications
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Towards an Automated Waste-Sorting System: Computer Vision and Neural-Network Inverse Kinematics for a 5-DOF Arm

Dao-Thi Mai-Phuong, Sy-Dung Vu, Nguyen Phan-Duy-Tan, Tran-Thi Kim-Thanh et al.
Zenodo (CERN European Organization for Nuclear Research)
Advanced Neural Network Applications
article

Towards an Automated Waste-Sorting System: Computer Vision and Neural-Network Inverse Kinematics for a 5-DOF Arm

Dao-Thi Mai-Phuong, Sy-Dung Vu, Nguyen Phan-Duy-Tan, Tran-Thi Kim-Thanh, N.-K. Nguyen, Nguyen Duc-Hieu
article en

Abstract

Automated waste sorting with a robotic manipulator requires two capabilities that belong to different fields: recognizing and localizing the object in the image, and converting a target Cartesian position into joint angles. This paper reports the construction and separate evaluation of both modules for a 5-DOF manipulator, and verifies the inverse-kinematics module on a continuous trajectory. For the recognition module a dedicated dataset, EPU_waste, of 2,955 images and 4,068 annotated instances in three classes was built, and two instance-segmentation models, YOLOv8s-seg and YOLO11s-seg, were trained and compared on it. YOLOv8s-seg reaches a recall of 0.97 against 0.88 and runs at 56.5 frames per second on a Quadro RTX 3000 GPU, and was therefore selected. For the inverse-kinematics module a 7–15–15–4 feedforward network with 424 weights was trained by the Levenberg–Marquardt algorithm over ten independent runs, giving a joint angle error of 2.2164 ± 0.1811° and a position error of 9.96 mm. On a continuous circular trajectory, the network gives a position RMS error of 3.461 mm with a standard deviation of only 1.30 mm, so the error is largely a near-constant systematic offset.

Zenodo (CERN European Organization for Nuclear Research)
Electric Power University (VN), Hanoi University of Industry (VN)
Advanced Neural Network Applications
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon