AD-MPCC: Adaptive Differentiable Model Predictive Contouring Control for Autonomous Racing

AD-MPCC integrates differentiable Model Predictive Contouring Control (MPCC) with online Pacejka tire-parameter estimation for autonomous racing under changing road-surface conditions. The controller adapts both the tire model and the MPCC objective weights online, enabling stable high-speed racing when the friction profile changes across the track.

Sources: arXiv paper | source code | IROS 2026 | F1TENTH Gym

Paper

AD-MPCC: Adaptive Differentiable Model Predictive Contouring Control for Autonomous Racing

Nam T. Nguyen, Binh Nguyen, Ahmad Amine, Thanh Vo-Duy, Rahul Mangharam, and Truong X. Nghiem.

Method Overview

AD-MPCC combines three components at each control step:xfdow

• Online parameter estimation: a prior-regularized moving-horizon estimator with exponentially decaying weights updates Pacejka tire parameters to track surface transitions in real time.
• Differentiable MPCC: implicit-function-theorem sensitivities of the MPCC solution with respect to objective weights identify high-performing MPCC weight choices.
• PaIML adaptation: a Pacejka-informed machine learning model maps the estimated tire/surface state to MPCC weights for real-time control.

Simulation Figures

Single-Surface Scenario

All controllers are evaluated on a uniform surface with mu_max = 1.2 for 10 laps. Diff-MPCC and AD-MPCC reduce lap time by about 11 seconds compared with the fixed-weight baseline, while AD-MPCC gives the smallest mean lateral offset.

Multi-Surface Scenario

The multi-surface race varies from mu_max = 0.7 to mu_max = 1.2 across different track sections. In the reported experiments, AD-MPCC is the only controller that completes the lap; MPCC, A-MPCC, and Diff-MPCC crash at the first low-friction segment.

Citation

@misc{nguyen2026admpccadaptivedifferentiablemodel,
      title={AD-MPCC: Adaptive Differentiable Model Predictive Contouring Control for Autonomous Racing},
      author={Nam T. Nguyen and Binh Nguyen and Ahmad Amine and Thanh Vo-Duy and Rahul Mangharam and Truong X. Nghiem},
      year={2026},
      eprint={2607.00141},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2607.00141},
}