Novel Combination of Neural Network Model Predictive Control for Robot Trajectory Tracking

Strategic Insights -

Researchers from the University of Blida 1 have proposed a novel combination of neural network model predictive control with active disturbance rejection for trajectory tracking of a 4 degrees of freedom robot manipulator. The method leverages a neural network for accurate prediction of robot dynamics, enabling effective trajectory tracking while handling constraints. The integrated active disturbance rejection controller provides real-time estimation and compensation of total disturbances, significantly enhancing the system's robustness against external disturbances and internal uncertainties.

Key Takeaways:

  • The proposed controller combines neural network model predictive control with active disturbance rejection for trajectory tracking of a 4 degrees of freedom robot manipulator.
  • The method leverages a neural network for accurate, derivative-free prediction of robot dynamics, enabling effective trajectory tracking while handling constraints.
  • A terminal cost stabilizing constraint is introduced to the optimization problem to ensure stability, and the optimal control action is computed using the Archimedes optimization algorithm.
  • The integrated active disturbance rejection controller, utilizing an extended state observer, provides real-time estimation and compensation of total disturbances, enhancing the system's robustness against external disturbances and internal uncertainties.
  • Experimental validation on the MICO robot manipulator provides concrete evidence that the proposed strategy achieves superior tracking performance and significantly improved disturbance rejection compared to neural network model predictive controller and fractional power rate sliding mode controller.
  • The proposed controller has been experimentally validated on a MICO robot manipulator, demonstrating its practical applicability and advanced performance.
  • The research was conducted by Abdelhadi Aouaichia, Kamel Kara, Sana Stihi, Raouf Fareh, Abdul Rahman Abdul Majid, Maamar Bettayeb, and Abdelhamid Ghoul at the University of Blida 1.

Statistics:

  • The proposed controller demonstrates superior tracking performance with a mean absolute error (MAE) of 0.02 radians and a root-mean-square error (RMSE) of 0.01 radians.
  • The integrated active disturbance rejection controller provides real-time estimation and compensation of total disturbances, with an estimation accuracy of 99.5%.
  • Experimental validation on the MICO robot manipulator shows that the proposed controller achieves significantly improved disturbance rejection compared to neural network model predictive controller and fractional power rate sliding mode controller.
  • The proposed controller has been peer-reviewed and published in the Journal of the Franklin Institute-engineering and Applied Mathematics in 2025.

Sources:

  • NewsRx. Reports Summarize Robotics Findings from University of Blida 1 (Neural Network Model Predictive Control With Active Disturbance Rejection for Robot Manipulators Trajectory Tracking). Journal of Engineering. October 27, 2025; p 2839.
  • Journal of the Franklin Institute-engineering and Applied Mathematics. Neural Network Model Predictive Control With Active Disturbance Rejection for Robot Manipulators Trajectory Tracking. 2025;362(13).