Mission Performance: Automatic and Adaptive Race Pace Progression for Autonomous Racing

I. INTRODUCTION

• Autonomous races require vehicles to manage performance without human intervention, while adapting to changing grip and validating stability as they approach handling limits.The progression aims to approach those limits in the shortest possible time while dynamically scaling performance down when necessary.Source: §I · p. 1
• The Mission Performance module manages lateral, longitudinal, and combined performance using tire warm-up progression, real-time safety metrics, and a temperature-based lookup table.Performance is limited as needed to maintain stability.Source: §I · p. 1

II. RELATED WORKS AND CONTRIBUTION

• Prior motorsport strategy work focuses on performance optimization, including lap-time minimization and energy/fuel consumption, while generally assuming a proficient human driver handles vehicle dynamics such as stability under varying conditions.Examples include hybrid power-unit fuel minimization, electric-endurance race-distance maximization, and single-lap time optimization.Source: §II · pp. 1–2
• Online model adaptation has been demonstrated on a Formula Student prototype, but its validation was at relatively low velocities and high-speed model updates may risk instability.The stated concern is especially relevant to optimization-based algorithms operating near handling limits.Source: §II · p. 2
• The authors describe this as the first work, to their knowledge, to manage a complete autonomous race from warm-up while adapting performance to track conditions in real time.The cited works include some online adaptation, but do not explicitly target an increasing progression based on safety and vehicle stability.Source: §II · p. 2

III. SYSTEM OVERVIEW

• A2RL-format races involve up to six vehicles competing without human intervention, with Race Control managing on-track operations through digital flags.A YELLOW flag limits speed to 80 km/h when an incident occurs.Source: §III · p. 2
• The overview frames performance progression within a race that transitions from pit-lane departure to the return at the run’s end.Race Control signals the start of unlimited-speed racing with a GREEN flag after the group forms and the lead car crosses the start/finish line.Source: §III · p. 2
• The architectural subset centers on the Mission Performance module and its interacting components.The overview omits implementation details for external modules, focusing on aspects affected by automatic performance progression.Source: Figure 1 · p. 2, §III · p. 2

A. Grip Estimation

• Grip Estimation reconstructs the tire force-versus-slip relationship and identifies Pacejka Magic Formula parameters through nonlinear least squares.The relationship is reconstructed for the front and rear axles from estimated slips, lateral forces, and vertical loads.Source: A. Grip Estimation · pp. 2–3
• Dyf and Dyr converged to approximately 1.66 and 1.76 from initial values of 1.75 after estimated friction deviated by more than 5% around 450 s.The identified values indicate a more understeering balance, consistent with tire mileage exceeding 250 km.Source: Figure 2 · p. 2, A. Grip Estimation · p. 3
• As tire temperature rises, the estimated peak lateral grip rises; Figure 2 plots axle grip over time alongside average axle tire temperature.The temperature is measured from the inner tire core using TPMS sensors.Source: Figure 2 · p. 2, A. Grip Estimation · p. 3

B. Mission Performance

• Mission Performance independently tunes lateral grip, acceleration, braking, and combined tire utilization to regulate the vehicle operating point through a normalized friction ellipse.The targets are high-level inputs to the longitudinal planner.Source: B. Mission Performance · p. 3, Figure 3 · p. 3
• In target-scaling mode, tire parameters remain fixed while normalized temperature-dependent gains scale performance targets; this avoids continuously modifying model parameters in the optimization-based planner and controller.The temperature–performance map can be identified offline from experimental data, and the target-scaling mode was selected for competition.Source: B. Mission Performance · p. 3

• The planner computes pathwise velocity references by forward–backward integration, constraining admissible speed by path curvature and maximum longitudinal and lateral accelerations.Source: C. Longitudinal Planner · p. 3
• Maximum longitudinal and lateral forces are derived from available tire grip, vehicle mass, and vertical tire loads through the stated friction law.Source: C. Longitudinal Planner · p. 3
• A generalized friction ellipse constrains admissible longitudinal and lateral accelerations under combined-slip conditions.Source: C. Longitudinal Planner · p. 3
• Mission Performance tunes the ellipse exponent, using separate longitudinal and lateral exponents to shape combined-slip behavior independently along each axis.Source: C. Longitudinal Planner · pp. 3–4

• The MPP optimizes the global path and longitudinal speed profile while accounting for deviations from the nominal trajectory.Source: D. Model Predictive Planning and… · p. 4
• The MPP provides a high-fidelity reference for the MPC.Source: D. Model Predictive Planning and… · p. 4
• The MPP and MPC share the same vehicle model and parameters, ensuring consistency between planning and control.Source: D. Model Predictive Planning and… · p. 4

A. Track Sectors

• Track sectors follow the Frenet coordinate of a reference line and cover turns or straights where performance changes can help minimize lap time while maintaining safety.Source: A. Track Sectors · p. 4
• Figure 4 maps the performance sectors on Yas Marina's north layout; a separate low-performance pit-lane profile applies at pit-out and near pit entry.Source: Figure 4 · p. 4, A. Track Sectors · p. 4
• Sector-based updates can reduce performance only in a locally slippery or unstable section, preserving maximum pace in unaffected sections rather than limiting the whole lap to the worst-case corner.Source: A. Track Sectors · p. 4

• Metrics are computed at 20 Hz and include lateral tracking error, understeer or oversteer angle, longitudinal acceleration and slip, and TC/ABS activations.A 10-sample moving average at 20 Hz gives a 0.5 s window for lateral error, understeer degree, and lateral velocity estimates to reduce transient-triggered decreases.Source: B. Logic and Implementation · p. 4
• Performance updates are skipped in the pit lane, during warm-up laps, and when a race-control flag limits vehicle speed.Source: B. Logic and Implementation · p. 4
• The managed parameters include lateral and longitudinal acceleration or braking capability and the longitudinal force sustained while cornering, affecting corner-exit traction and trail-braking aggressiveness.Source: B. Logic and Implementation · p. 5
• Performance is reduced by half the prior increment above a metric threshold, held between 90% and 100% of the threshold, and increased one step below 90%.Source: B. Logic and Implementation · p. 5
• The safety hierarchy requires Ky to exceed a sector-tuned threshold before n+ or n− can increase, and permits reducing Ky only when n+ and n− are at minimum.Kx+ and Kx− can likewise be decreased only when n+ and n− are at minimum.Source: B. Logic and Implementation · p. 5
• Static per-parameter maxima are tuned from offline tests, while a front-tire-temperature LUT clamps safe lateral and longitudinal performance limits.The temperature clamp is intended to restore a safe baseline after sub-optimal thermal conditions, such as cooling during speed-limited periods.Source: B. Logic and Implementation · p. 5
• At each new track zone, the update loop uses completed-sector exceedance records, applies temperature and static clamps, adjusts performance, and resets local counters.Source: B. Logic and Implementation · p. 5
• A testing mode allows manual overrides of performance limits for selected sectors to support targeted testing of algorithm updates.Source: B. Logic and Implementation · p. 6

V. RESULTS

• The module was deployed during A2RL Season 2 at Yas Marina across Qualifying, Sprint Race, and Grand Finale sessions.The competition took place in Abu Dhabi in October and November 2025.Source: §V · p. 6
• Its core decision-making logic executes in under 0.2 ms on an Intel Core i9-14900HX processor.Resource-intensive grip and state estimation are delegated to upstream modules.Source: §V · p. 6
• The module is described as highly computationally efficient.The reported timing concerns its core decision logic, with grip and state estimation handled upstream.Source: §V · p. 6

A. Race Progression

• Performance parameters remained unchanged during the first two reconnaissance laps, then increased after the first full-speed lap.The car began increasing performance in T8 of lap 1 after receiving the GREEN flag before the start/finish line of its last reconnaissance lap.Source: A. Race Progression · p. 6
• Lateral performance increased uniformly across sectors, while combined performance increases depended on sector-specific lateral thresholds.Those thresholds varied by sector.Source: A. Race Progression · p. 6
• A lapping maneuver on lap 6 triggered understeer protection and reduced combined performance while the car yielded the racing line.The maneuver required at least 3.5 meters of clearance from the track boundary; the trailing vehicle remained within 15 meters, extending the yield for four laps.Source: A. Race Progression · p. 6
• The performance reduction after an erroneous IMU reading demonstrated autonomous management of unforeseen state-estimation anomalies during a race.The reading lasted approximately half a second and produced inaccurate longitudinal and lateral velocity estimates.Source: A. Race Progression · p. 6

B. Oversteering Scenario

• Oversteering metrics triggered performance reductions in T1 on laps 15, 16, and 17 after targets had followed nominal progression through lap 8 and remained at maximum clamps through lap 14.The scenario occurred during the final three laps on an old tire set, before the run ended with a pit-in request.Source: B. Oversteering Scenario · p. 6
• The reduction brought the performance value close to the threshold, but the vehicle remained not fully safe in stability.The authors identify the reduction step and thresholds as insufficiently conservative in this case.Source: B. Oversteering Scenario · p. 6
• The case highlights the system’s dependence on tuning the reduction step and safety thresholds.The response was triggered by oversteering metrics during the final three laps on old tires.Source: B. Oversteering Scenario · p. 6

VI. CONCLUSION AND FUTURE WORK

• The module dynamically adapts longitudinal, lateral, and combined performance parameters per sector, automating warm-up and responding to track and vehicle conditions without human intervention.The paper characterizes it as a robust solution for managing the trade-off between vehicle performance and safety.Source: §VI · p. 6
• Future work calls for tighter planner integration to handle multi-agent yields and overtakes.Explicit planner signals for evasive or overtaking maneuvers could let the module discard isolated data rather than treating it as a clear dynamic limit, avoiding unwarranted performance reductions.Source: §VI · p. 6
• The authors aim to replace static progressions with a tire-temperature-driven exploratory LUT bounded to remain safe while allowing the vehicle to push its limits.The planned approach would require selecting high-level aggressiveness profiles tailored to vehicle conditions, such as new versus used tire compounds.Source: §VI · p. 6
• The authors also plan to explore a hybrid mode that updates model parameters when actual mismatch exceeds a threshold.This is described as future work involving closer interaction with the Grip Estimation module.Source: §VI · pp. 6, 8
• The conclusion’s cited visuals include a lap-by-lap, per-sector performance progression table and figures on safety metrics and T1 oversteer.The supplied captions identify these visuals’ scope but do not provide their underlying values.Source: Table II · p. 7, Figure 5 · p. 7, Figure 6 · p. 8