Real-time position and orientation determination of multiple robots in Robocup soccer using lightweight deep learning models

Özkurt A., Öztekin A. E.

SCIENTIFIC REPORTS, cilt.1, sa.1, ss.1-20, 2026 (SCI-Expanded, Scopus)

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 1 Sayı: 1
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1038/s41598-026-65913-7
  • Dergi Adı: SCIENTIFIC REPORTS
  • Derginin Tarandığı İndeksler: Academic Search Ultimate (EBSCO), Natural Science Collection (ProQuest), Biological Science Database (ProQuest), Biomedical Reference Collection: Corporate Edition (EBSCO), Health Research Premium Collection (ProQuest), Scopus, Science Citation Index Expanded (SCI-EXPANDED), BIOSIS, Chemical Abstracts Core, EMBASE, MEDLINE, Directory of Open Access Journals, Zoological Record
  • Sayfa Sayıları: ss.1-20
  • Dokuz Eylül Üniversitesi Adresli: Evet

Özet

Real-time detection of robot position and orientation is a critical challenge in multi-robot autonomous soccer systems. This paper presents two complete algorithms for position and orientation determination of multiple RoboCup robots from overhead camera images. In the first algorithm, blob analysis with HSV color segmentation is used for position detection, and two newly designed lightweight CNN architectures (RoboCup-I and RoboCup-II) are used for orientation estimation of single-color robots. In the second algorithm, SSD-MobileNetV2 and RFB-ULGFD object detection models are compared for robot and ball position detection, while a novel color segmentation-based trigonometric orientation algorithm is proposed for multi-robot orientation detection. Experimental results show that the RFB-ULGFD model with 480×360 input achieving a mean Average Precision (mAP) of 98.17% at IoU 0.5 with a model inference time of only 4 ms on GPU (12 ms including preprocessing and post-processing within the full pipeline). The proposed color-segmentation-based orientation algorithm achieves a Mean Squared Error (MSE) of 2.06 degrees on synthetic data with a total pipeline latency of only 6–15 ms even in CPU environments, satisfying the demands of real-time multi-robot systems.