The relentless pursuit of the perfect lap
The virtual driver: motorsport’s ultimate weapon in the war for lap time
A single motorsport test campaign can cost millions of dollars. Documents reported in the McLaren court case involving Alex Palou indicated that Ryo Hirakawa paid approximately USD $3.5 million for a package that included one Formula 1 free practice session and two preparatory tests with McLaren.
At that level of investment, simulation accuracy becomes a business-critical issue. Yet one of the most influential contributors to lap time remains among the most difficult to simulate accurately: the driver. Modern race cars combine hybrid powertrains, active aerodynamics, advanced suspensions and tightly coupled controls, but lap time ultimately depends on how a driver exploits that system in real time.
Driver behavior is not a fixed input. Braking, turning, throttle application and correction strategies continuously adapt to vehicle response, grip, balance and track conditions. What if an inaccurate driver model sent your team to the track with the wrong setup after investing millions in testing? In modern motorsport, a single test campaign can cost millions of dollars. Yet many simulation workflows still rely on virtual drivers that follow predefined references instead of actively optimizing performance.
If the virtual driver model is wrong, every prediction built on it (lap time, tire use, energy consumption and setup sensitivity) becomes unreliable and a business-critical risk.
The race car: a system of systems
A race car is a system of systems: propulsion, energy storage, power electronics, cooling, gearbox, chassis, tires and aerodynamics interact continuously. Optimizing one domain in isolation misses the real performance problem.
The power unit creates both propulsion and heat. The gearbox and axles transmit torque while adding losses and loads. Power electronics manage energy delivery and recovery, while the energy storage system defines the usable performance envelope. Thermal management trades cooling capacity against mass and drag. Chassis, suspension and tires convert aerodynamic and mechanical loads into grip, while aerodynamics governs downforce, drag and stability.
At the center is the virtual driver: the adaptive system translating all vehicle behavior into lap time. If the driver model cannot respond like a skilled virtual driver, the engineering conclusions drawn from simulation are compromised.
Why system simulation is the answer
The engineering challenge is to predict and optimize a vehicle whose subsystems interact at their limits under changing track and environmental conditions.
Physical testing remains essential, but it is expensive, limited and unable to explore the full design space. A test validates one configuration under one set of conditions; it cannot systematically isolate subsystem effects or predict untested scenarios.
System simulation fills that gap by connecting powertrain, thermal, electrical, mechanical, aerodynamic, tire and control behavior in one multi-physics digital twin. Engineers can:
- Explore many setup and design alternatives rapidly
- Isolate subsystem contributions to total performance
- Predict behavior in extreme or untested conditions
- Optimize across the full vehicle rather than one domain at a time
- Validate controls before real hardware deployment
The result is faster development, higher-confidence decisions and setup choices validated against a complete vehicle model before track time begins.
The race car is a system of systems. System simulation is how you master it.
What the market offers today, and where it falls short
Current virtual driver modeling approaches each solve part of the problem, but none fully combines optimization, adaptation, integration and real-time capability.
1. PID-based closed-loop tracking virtual drivers
PID drivers use feedback control to reduce error against a predefined trajectory/speed profile.
Strengths
- Accurate tracking of prescribed speed and line
- Reliable execution of standard maneuvers
- Low computational cost and real-time operation
Limitations
- Tracks rather than optimizes
- Does not adapt naturally to hardware or conditions
- Reactive, with limited look-ahead intelligence
- Often requires manual tuning after changes
2. Quasi-static lap simulation tools
Quasi-static tools estimate maximum speed around a circuit from steady-state vehicle limits.
Strengths
- Fast first-order lap-time estimates
- Low model complexity
- Useful early in design trade-off studies
Limitations
- Miss transient dynamics, weight transfer, tire evolution and suspension behavior
- Cannot capture driver-vehicle interaction
- Produce theoretical limits rather than realistic driven laps
3. Human-in-the-loop driving simulators
Human-in-the-loop simulators use a real driver with a high-fidelity vehicle and track environment.
Strengths
- Capture real human adaptation and feedback
- Support final sign-off and driver confidence
- Provide subjective insight no algorithm can fully replicate
Limitations
- Require scarce expert driver time
- Cannot run faster than real time
- Produce variable results across sessions
- Cannot explore thousands of configurations efficiently
The missing capability is a virtual driver that optimizes, adapts and integrates with the full vehicle model while remaining fast enough for practical engineering use.
A new philosophy: The driver model as a human digital twin
The solution is not simply a better tracking controller. It is an intelligent optimization agent embedded in the complete vehicle digital twin.
The Simcenter Amesim Advanced Virtual Driver Model, developed by the Simcenter Engineering Services team, operates in closed loop with the race car digital twin. Instead of following a fixed reference, it computes the best control strategy at each instant, with the virtual driver and vehicle co-simulated as one coupled system.
The architecture: NMPC at the heart
At its core is Nonlinear Model Predictive Control (NMPC). At each time step, the controller solves a constrained optimization problem over a prediction horizon, looking ahead to determine steering, throttle and braking inputs.
The formulation balances three objectives: staying close to the reference trajectory, minimizing energy use and reducing lap time over the upcoming horizon, allowing the virtual driver to deviate from the reference. Because racing line and speed profile are optimized together, the result respects vehicle limits in terms of stability and safety while searching for the true performance opportunity.
Using the RESAFE/COL method, the NMPC achieves about 10X speed-up versus state-of-practice MPC methods and can generate a standalone library deployable on target hardware.
The driver model also satisfies a rich set of physical and operational constraints simultaneously:
- Vehicle physical evolution at each time step
- Steering, braking, throttle and force limits
- Velocity, acceleration and slip constraints
- Track boundaries and obstacle avoidance
The result is look-ahead optimization within a strict real-time CPU budget.
What this means: The virtual driver is no longer reproducing a lap. It is actively searching for a faster one.
Adapting to any vehicle, any track, any condition
Optimization alone is not enough if every new setup requires manual tuning. The Simcenter Engineering Services approach adds controller autocalibration.
The framework tunes NMPC cost weights and driver parameters using performance outcomes such as lap time and stability. Parallel exploratory digital twins compare candidate parameter sets and converge automatically for the current vehicle, track and conditions. In validation, autocalibration extracted up to 30 seconds from the lap versus the initial reference.
What this means: Engineers no longer need to manually retune the driver for every vehicle setup or circuit.
From wet weather to overtaking: Real world versatility
The real value appears when conditions depart from the ideal.
Wet weather adaptation: When grip changes, the NMPC driver adapts trajectory and speed to the new envelope. In validation, it achieved a 161.48-second wet lap by finding a better reduced-grip line rather than simply slowing the dry reference.
Overtaking and traffic scenarios: Obstacle constraints are built into the NMPC architecture, enabling overtaking simulations at racing speed, including boosted and DRS-assisted scenarios above 330 km/h.
What this means: The virtual driver adapts to changing grip levels the same way a skilled racing driver would.
The digital twin foundation
The Advanced Virtual Driver Model depends on a high-fidelity vehicle digital twin. It receives real-time feedback on dynamics, tires, powertrain and aerodynamics, then returns optimized control inputs for the next simulation step. The driver shapes the car’s behavior, and the car shapes the driver’s next decision.
The Simcenter Amesim race car digital twin includes:
- Powertrain – ICE, eMotor, gearbox, inverter, battery and thermal management
- Braking – front/rear systems with regenerative braking
- Suspension and chassis – multi-body approach with physical damper models
- Tires – MF-Swift with rigid ring dynamics
- Aerodynamics – variable drag/lift coefficients and active surfaces
- 3D track environment – road import, trajectory setup, G-G diagrams and non-flat terrain features
This is a complete multi-physics closed loop: the NMPC driver optimizes against the full digital twin, not a simplified speed demand.
What this means for engineering teams
For motorsport and performance engineering teams, the impact is direct:
- Faster development: evaluate many configurations faster than human-in-the-loop simulator.
- Higher-confidence: base lap-time predictions on physical optimization, not fixed references.
- Seamless adaptation: automatically recalibrate for vehicle, track and condition changes.
- Real-time capability: use the same architecture offline, in real time and on embedded targets.
Conclusion: The driver model has grown up
For years, the virtual driver was a necessary approximation. The Advanced Virtual Driver Model developed by the Simcenter Engineering Services team and deployed in closed loop with the Simcenter Amesim race car digital twin changes that role.
By combining NMPC optimization, autocalibration, full multi-physics integration and real-time deploy-ability in an open FMI-compatible architecture, it delivers a driver model that does more than follow the car: it drives it toward the limit.
At $3.5 million per test package, the cost of getting it wrong has never been higher. With Simcenter Amesim, teams can arrive at the track with stronger decisions already validated virtually.
The race car is a system of systems. The driver model must be worthy of that complexity. With Simcenter Amesim, it finally is.
[1] https://www.sciencedirect.com/science/article/pii/S240589632401437X
Want to try Simcenter Amesim yourself?
