A Comprehensive Benchmark of Vehicle Dynamics Models for Autonomous Racing: a Deep Dive into MPC

This work benchmarks vehicle dynamics models commonly used in autonomous racing, presenting their structures, key parameters, and simplifying assumptions. We analyze their applicable ranges by com paring solutions proposed in the literature and highlighting limitations through simulations, including both standard and real on-track maneu vers, where differences in accuracy and predictive capability are evalu ated. A validated methodology for parameter identification is also pre sented, alongside a discussion of numerical stability and model lineariza tion, which are critical for real-time optimal control applications. In the f irst part of this work, through open-loop simulations, simplified mod els are evaluated against a high-fidelity multi-body model, focusing on vehicle dynamics accuracy and integrator stability. In the second part, the models are integrated into a Model Predictive Controller (MPC) to assess their effectiveness in closed-loop path-tracking within an au tonomous racing context. Real-time simulations are employed, with the multi-body model serving as physical representation. Experiments with a Dallara EAV24 Super Formula at the Yas Marina circuit validate the ap proach, demonstrating its practical applicability and further supporting the findings.