...

KINETIQA

Kinematic Integration Network with Enhanced Temporal Intelligence and Quality-driven Attention

Lyes Saad Saoud, Irfan Hussain
Khalifa University, ARIC Center
arXiv Code 🤗Checkpoints Supplementary Video

Real-time knee joint angle prediction for exoskeleton-based gait rehabilitation.

Abstract

We present KINETIQA, a Transformer-based model tailored for real-time knee joint angle prediction in exoskeleton-assisted rehabilitation. It integrates temporal convolutions, recursive feature elimination, and quality-driven attention to enhance long-horizon gait forecasting while maintaining computational efficiency. KINETIQA significantly outperforms baselines like Transformer, TempoNet, LSTM, and FocalGatedNet, achieving over 20% MAPE reduction and demonstrating resilience to various sensor noise types including Gaussian, uniform, and salt-and-pepper noise. Inference latency remains under 25ms on RTX 4090 hardware, making it suitable for closed-loop biomechanical control in real-world settings.

BibTeX

@article{saoud2025kinetiqa,
  title={KINETIQA: Kinematic Integration Network with Enhanced Temporal Intelligence and Quality-driven Attention for Precise Joint Angle Prediction in Exoskeleton-based Gait Analysis},
  author={Saad Saoud, Lyes and Hussain, Irfan},
  journal={Submitted to IEEE Access},
  year={2025},
  note={Under review}
}

Reproducibility Statement

To ensure transparency and support future research, we have created a dedicated project webpage at https://github.com/LyesSaadSaoud/KINETIQA/. The page provides comprehensive access to all resources required to reproduce our experiments, including:

  • Source code and trained model checkpoints
  • Visualization animations and result plots
  • Detailed step-by-step instructions for training and evaluation
  • Hardware and software environment details
  • Downloadable requirements.txt and configuration files
  • Input/output data formats and expected folder structures

All experiments were implemented using PyTorch 1.13.1 in PyCharm (Community Edition 2023.1) on a Windows 11 system with an NVIDIA RTX 4090 Ti GPU, CUDA 11.7, and cuDNN 8.6. Python version 3.10.9 was used, and environment setup was managed using venv and pip. While Docker was not used, the complete environment and setup instructions are thoroughly documented on the project page to ensure replicability.