Neural Network and ANFIS based auto-adaptive MPC for path tracking in autonomous vehicles

Kebbati, Yassine; Ait-Oufroukh, Naima; Vigneron, Vincent; Ichalal, Dalil · 2021 · Crossref

DOI: 10.1109/icnsc52481.2021.9702227

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of maintaining robust lateral control in autonomous vehicles operating under varying environmental conditions and external disturbances. Standard Model Predictive Control (MPC) strategies often fail to handle uncertainties such as changing road adhesion, wind gusts, and variable longitudinal velocities. To overcome these limitations, the authors propose an auto-adaptive MPC controller that dynamically adjusts its prediction model parameters in real-time. The motivation stems from the need for a control system that preserves safety constraints and tracking accuracy despite the constantly changing working conditions inherent to self-driving scenarios. The methodology involves a two-stage process: offline optimization and online adaptation. First, an improved Particle Swarm Optimization (PSO) algorithm is employed to tune and optimize key MPC parameters—specifically the prediction horizon ($N_p$), control horizon ($N_c$), and weighting matrices ($Q$ and $R$). This optimization accounts for a wide range of longitudinal speeds, lateral references, road adhesion coefficients, and lateral wind disturbances, generating a dataset of optimal parameters. Second, two distinct learning approaches are used to map current vehicle states to these optimal parameters for online adaptation: Multi-Layer Perceptron (MLP) Neural Networks and Adaptive Neuro-Fuzzy Inference Systems (ANFIS). The vehicle dynamics are modeled using a single-track dynamic bicycle model, while the evaluation utilizes a high-fidelity 3-DOF dual-track model with nonlinear tire characteristics in MATLAB. The results demonstrate that both adaptive controllers significantly outperform standard MPC in terms of tracking accuracy and disturbance rejection. In a triple lane change scenario involving wind gusts and varying road adhesion, the Neural Network-based MPC (NN-MPC) achieved the lowest Mean Squared Error (MSE) of 0.0051, compared to 0.0062 for ANFIS-MPC and 0.0318 for standard MPC. In a general trajectory tracking test, NN-MPC again showed superior accuracy with an MSE of 0.132, versus 0.1349 for ANFIS-MPC and 0.5896 for standard MPC. While ANFIS-MPC produced smoother control signals, it suffered from the curse of dimensionality, making it computationally demanding and less practical for real-time applications compared to the neural network approach. The significance of this work lies in demonstrating that integrating AI-based parameter adaptation with MPC enhances the robustness of autonomous vehicle lateral control. The proposed NN-MPC controller effectively handles external disturbances and varying working conditions without the computational burden associated with fuzzy logic systems. This approach provides a viable solution for improving the reliability of autonomous driving systems in unpredictable environments, offering a clear advantage over static control strategies.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• lane positioning

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Theoretical Contribution: computational model