Putting the driver in the simulation: sim racing and driver-in-the-loop
Offline lap simulations assume a driver at the limit everywhere. Driver-in-the-loop simulators add the human – their perception, confidence and technique – and all the data-analysis methods transfer one-to-one to virtual laps.
Putting the driver in the simulation: sim racing and driver-in-the-loop: L1 Foundation explanation
Why add the driver? Most lap simulations have no driver model: the car is driven at the grip limit at every point. A setup can therefore be theoretically fastest yet undriveable – the driver can't exploit it because the car's feedback (steering torque, yaw cues, vibrations, sounds) doesn't let them operate confidently at the limit.
Driver-in-the-loop (DIL) runs a real-time vehicle model driven by a real driver's steering, throttle, brake and gear inputs, with visual, audio and (optionally) motion and force-feedback cues.
Levels:
- Sim racing platforms (e.g., iRacing, rFactor Pro-based and similar): laser-scanned tracks, licensed car models, data export to MoTeC i2 or ATLAS. Segers notes that all the analysis techniques in his book apply one-to-one to sim data.
- Professional static and motion simulators: team-specific vehicle models, correlated to the real car; used by F1, Hypercar, IndyCar, Super Formula and many GT and junior-formula teams, or rented from specialist centres.
Uses: driver preparation for new tracks, engineer training (including strategy scenarios), testing data-analysis methods and math channels before buying sensors, and generating data for track models where no real data exists.
Putting the driver in the simulation: sim racing and driver-in-the-loop: L2 Practitioner explanation
What DIL is good for.
- Track learning: brake points, gears, references, lines – especially valuable for junior drivers before a new circuit.
- Setup direction screening: does the driver prefer more rear stability on entry, can they exploit a more aggressive aero balance? Drivers rank options; the team brings the 2–3 most promising to the track.
- Procedures: starts, pit-lane entry/exit, energy-deployment modes, steering-wheel switch changes, safety-car restarts.
- Engineer training: running a virtual session or race with run plans, radio calls and strategy decisions.
What it struggles with.
- Sustained acceleration can't be reproduced; motion platforms "trick" the vestibular system with tilt (using gravity to simulate sustained acceleration), washout filters and onset cues. Tilt coordination only works for cues below 1 g, real platforms tilt at most ~20–30° (≈ 0.35–0.5 g), and the tilt must be applied below the rotation-perception threshold (~3 deg/s) or the driver feels the rotation – reaching 25° takes at least ~8 s – longer than most braking zones. Sustained cues are therefore scaled down and delayed; onset cues and visual flow carry the rest. Drivers report the weakest realism in long high-g corners and heavy braking.
- Tyre behaviour – thermal and wear models are the least mature part of most models; degradation runs are limited.
- Latency – total cue latency (input → model → visual/motion) needs to be low (tens of ms) or drivers over-correct.
Analysing sim data. Treat it exactly like car data: overlays, delta time, braking and throttle metrics, steering smoothness. Correlate sim vs real for the same driver: speed traces and lateral g usually agree well; the largest differences tend to appear in low-speed corners (tyre model) and in chassis inertial motions – Segers' Corvette C6R example matched speed and lateral g closely, with suspension high-frequency content (bumps from a scanned surface) matching remarkably well.
Putting the driver in the simulation: sim racing and driver-in-the-loop: L3 Advanced explanation
Correlating a DIL simulator.
- Objective correlation: drive the real reference setup; compare per-corner minimum speeds, braking points, gear usage, steering angles and throttle traces with real data from the same driver. Targets: lap time within a few tenths, minimum speeds within a few km/h, steering angles within ~10 %.
- Subjective correlation: does the driver feel the same balance changes in the sim as on track for known setup changes (e.g., one ARB step)? A sim that is objectively close but subjectively different is of limited use for setup work.
- Sensitivity correlation: setup changes should produce the same direction and magnitude of lap-time and balance changes as on track. This matters more than absolute lap time.
Pitfalls.
- Drivers adapt to the sim: they learn its quirks and become "sim-fast"; compare their technique metrics with their real-car metrics to detect sim-specific driving.
- Over-reliance: a setup tuned in the sim may hide a model error (e.g., tyre temperature effects). Always validate with targeted track running.
- Confidence effects: no consequences for crashing changes risk-taking – useful for exploring limits, misleading for judging how a driver will behave in a real car.
Bringing offline and DIL together. A typical modern workflow: offline (QSS or optimal-control) sims explore the setup space → DIL evaluates driveability and driver preference for the shortlisted options → the track validates. The engineer's job is to keep the three consistent and to log what each tool predicted versus what happened.
Key formulas
- platform tilt angle (valid only for a_cue < g; practical limit ~20–30°, applied below ~3 deg/s)
- sustained acceleration cue actually rendered (m/s²) – a scaled-down fraction of the car’s 2–5 g
Trackside takeaways
Offline sims assume a perfect driver; DIL tests whether a real driver can exploit a setup.
Sim data can be analysed with exactly the same tools as real data – use it to train engineers and develop math channels.
Correlate objectively (traces), subjectively (driver feel) and on sensitivities – not just lap time.
Motion platforms cannot reproduce sustained g; they rely on tilt, washout and onset cues.
Watch for drivers adapting to sim quirks; validate sim-derived setups on track.
On your car
Limited testing makes DIL essential for track learning and setup screening; many teams use rented or in-house simulators before each round.
Check yourself
Why can an offline lap simulation recommend a setup that is slower in reality?
- Segers – Analysis Techniques for Racecar Data Acquisition — §15.6 Putting the Driver in the Simulation (§15.6.1 Sim Racing, §15.6.2 Motion-Based Driver-in-the-Loop Simulation)