Autonomous racing pushes planning and control algorithms to exploit vehicle performance at the handling limits while satisfying track and actuator constraints. This thesis develops an offline simulator-in-the-loop lap-time optimization framework for PoliMOVE Autonomous Racing Team. The objective is to generate a benchmark lap under the same simulation conditions used to test the team's online architectures. The case study is carried out on Yas Marina North, in the context of the Abu Dhabi Autonomous Racing League (A2RL), in which the team competes with the Dallara Eav-24. The framework is implemented in Simulink and couples a nonlinear optimization strategy with a high-fidelity simulator. At each optimization update, the current vehicle state and track data are used to solve a finite-horizon NLP, whose optimal wheel torques and steering are converted into throttle, braking pressure, and steering commands. The prediction model is a control-oriented planar double-track model formulated in curvilinear coordinates. It includes aerodynamics, quasi-static load transfer, combined-slip Pacejka tire forces, and motor saturation. The optimization is formulated as a free-trajectory problem: instead of tracking a predefined racing line or speed profile, the solver maximizes progress while enforcing track boundaries, actuator limits, slip-ratio bounds, and vehicle dynamics. A specific contribution is the introduction of the longitudinal slip ratio as an optimization state, coupled to wheel torque and longitudinal tire force through a quasi-static wheel-equilibrium constraint. This enables torque-based control inside the solver and accounts for braking and traction slip limitations without a downstream slip controller, while avoiding explicit wheel-speed dynamics. A sensitivity analysis is conducted on the main tuning parameters: slip-ratio, torque and steering rate penalties, and prediction horizon length. The selected configuration achieves a simulator lap time of 54.935 s, with coherent trajectory, actuation, speed and tire-utilization profiles. The proposed framework provides an offline benchmark for future comparison with online autonomous racing architectures.
La guida autonoma in pista spinge gli algoritmi di pianificazione e controllo a sfruttare il veicolo al limite dell'aderenza, rispettando i vincoli del tracciato e di attuazione. Questo lavoro sviluppa per PoliMOVE Autonomous Racing Team un'architettura offline simulator-in-the-loop per l'ottimizzazione del tempo sul giro, al fine di generare un riferimento nelle stesse condizioni di simulazione usate dal team per testare le proprie architetture online. Il caso studio riguarda una configurazione ridotta del circuito di Yas Marina, dove il team gareggia con la Dallara EAV-24 per l'Abu Dhabi Autonomous Racing League (A2RL). L'architettura, in Simulink, accoppia ottimizzazione non lineare e un simulatore ad alta fedeltà. Ad ogni iterazione, stato del veicolo e dati del tracciato definiscono un NLP a orizzonte finito; la soluzione fornisce la coppia alla ruota, convertita poi in acceleratore e pressione frenante, e lo sterzo. Il modello del veicolo nel solver è un double-track planare in coordinate curvilinee, comprensivo di aerodinamica, trasferimento di carico quasi-statico, modello pneumatici Pacejka combinato e saturazione del motore sul posteriore. L'ottimizzazione è a traiettoria libera: il solver massimizza l'avanzamento rispettando i limiti di pista, attuatori, slittamento longitudinale e dinamica del veicolo. Un contributo principale riguarda il rapporto di slittamento longitudinale (slip ratio), introdotto come stato di ottimizzazione e legato a coppia e forza longitudinale tramite equilibrio quasi-statico. Ciò permette il controllo in coppia nel solver e include i limiti di slittamento in frenata e trazione evitando la dinamica esplicita della ruota, e senza controllori a valle. L'analisi di sensitività valuta i parametri di tuning: penalità su slip ratio, variazioni di coppia e sterzo, e orizzonte di predizione. La configurazione scelta ottiene un tempo sul giro di 54,935 s, con profili coerenti di traiettoria, attuazione e velocità, fornendo un valido riferimento offline per architetture di guida autonoma.
Development of a simulator-in-the-loop lap-time optimization framework for autonomous racing
Cipriani, Filippo
2025/2026
Abstract
Autonomous racing pushes planning and control algorithms to exploit vehicle performance at the handling limits while satisfying track and actuator constraints. This thesis develops an offline simulator-in-the-loop lap-time optimization framework for PoliMOVE Autonomous Racing Team. The objective is to generate a benchmark lap under the same simulation conditions used to test the team's online architectures. The case study is carried out on Yas Marina North, in the context of the Abu Dhabi Autonomous Racing League (A2RL), in which the team competes with the Dallara Eav-24. The framework is implemented in Simulink and couples a nonlinear optimization strategy with a high-fidelity simulator. At each optimization update, the current vehicle state and track data are used to solve a finite-horizon NLP, whose optimal wheel torques and steering are converted into throttle, braking pressure, and steering commands. The prediction model is a control-oriented planar double-track model formulated in curvilinear coordinates. It includes aerodynamics, quasi-static load transfer, combined-slip Pacejka tire forces, and motor saturation. The optimization is formulated as a free-trajectory problem: instead of tracking a predefined racing line or speed profile, the solver maximizes progress while enforcing track boundaries, actuator limits, slip-ratio bounds, and vehicle dynamics. A specific contribution is the introduction of the longitudinal slip ratio as an optimization state, coupled to wheel torque and longitudinal tire force through a quasi-static wheel-equilibrium constraint. This enables torque-based control inside the solver and accounts for braking and traction slip limitations without a downstream slip controller, while avoiding explicit wheel-speed dynamics. A sensitivity analysis is conducted on the main tuning parameters: slip-ratio, torque and steering rate penalties, and prediction horizon length. The selected configuration achieves a simulator lap time of 54.935 s, with coherent trajectory, actuation, speed and tire-utilization profiles. The proposed framework provides an offline benchmark for future comparison with online autonomous racing architectures.| File | Dimensione | Formato | |
|---|---|---|---|
|
2026_07_Cipriani_Thesis.pdf
non accessibile
Descrizione: Thesis
Dimensione
8.33 MB
Formato
Adobe PDF
|
8.33 MB | Adobe PDF | Visualizza/Apri |
|
2026_07_Cipriani_Executive_Summary.pdf
non accessibile
Descrizione: Executive Summary
Dimensione
850.85 kB
Formato
Adobe PDF
|
850.85 kB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/259993