@article{4635,
  author = {Chunyan Zhao, Yinhui Hao},
  title = {Real-Time Trajectory Prediction for Robotic Networks Using Extreme Learning Machines},
  journal = {Journal of Networking Technology},
  year = {2026},
  volume = {17},
  number = {1},
  doi = {https://doi.org/10.6025/jnt/2026/17/1/18-27},
  url = {https://www.dline.info/jnt/fulltext/v17n1/jntv17n1_2.pdf},
  abstract = {This paper presents a comprehensive study on trajectory prediction using Extreme Learning Machines (ELMs),
with applications in autonomous systems, surveillance, and robotic networks. It begins by highlighting the
importance of time series and trajectory data in pattern recognition tasks, citing uses in visual surveillance,
human behavior analysis, and autonomous navigation. The paper reviews related work, emphasizing ELM's
advantages such as fast training, strong generalization, and avoidance of local minima over traditional
ANNs and SVMs. Specific studies are discussed, including LB-EBM for multimodal pedestrian prediction,
multilayer ELMs for video surveillance, and hybrid models like ELM-HTM for anomaly detection.
The core of the work details an ELM-based trajectory prediction algorithm, outlining input/output structures,
preprocessing steps, and offline/online phases. It distinguishes between single step, multi step, and rolling
prediction strategies, favoring the latter for balancing accuracy and error accumulation. The method is
validated on real world network robot data: 90 trajectories were collected, of which 70 were used for
training. Results show high prediction accuracy (e.g., X-axis errors 26.6 mm) and ultra fast inference (6 ms
per prediction), meeting real time robotic requirements. A spin classification experiment further demonstrates
ELM's efficacy, achieving 91.4% accuracy. The study concludes that ELM provides a lightweight, efficient,
and adaptable approach to real time trajectory forecasting in intelligent sports robotics and beyond.},
}