By ATN Editorial Team
A driver-in-the-loop (DIL) simulator is the verification tool that closes the simulation loop around a real human driver, in real time. The car, the tyres, the road, the traffic and the sensors are all virtual; the driver’s hands, feet, eyes and vestibular system are real, and the rig has to deliver visual, motion, audio and force-feedback cues at rates the human nervous system reads as continuous. DIL has been a core vehicle development tool in top-level motorsport for roughly two decades, and is increasingly important as a verification stage for ADAS and automated-driving features where driver behaviour forms part of the safety case. Regulators are increasingly incorporating simulation-based evidence into validation frameworks, although DIL evidence specifically is rarely a standalone homologation basis. This explainer covers what a DIL simulator actually is, how it differs from MIL, SIL, HIL and VIL, what the hardware and software stack looks like, the main vendor landscape, and where the technology is heading.
A DIL rig runs multiple tightly synchronised real-time subsystems in parallel. A vehicle dynamics solver — usually a multi-body or analytic plant model — typically runs at hundreds of Hz up to around 1 kHz on the real-time host, depending on model complexity and host hardware, computing chassis state from driver inputs and the virtual environment, with sub-loops handling tyres, brakes, electric power steering and powertrain. A motion-cueing algorithm converts the desired chassis accelerations into commands for the motion platform, exploiting tilt coordination and washout filtering to render long-duration cues inside a finite actuator envelope. A render pipeline produces visual frames typically at 60–120 Hz, with the actual rate dependent on GPU configuration and scene complexity. An audio engine produces engine, road and wind sound.
Transport delay — from driver input to perceived response — is the single most important system-level metric, but it has to be read carefully. Motion-platform vendors typically publish a sub-system latency measured from command issue to demanded acceleration on an IMU mounted on the platform; leading vendors quote figures below 5–10 ms on that basis. End-to-end perceived latency, which includes the visual pipeline, projector frame time and steering-feedback loop, is rarely published as a single measured number but generally sits in the low tens of milliseconds on tier-1 rigs and substantially higher on older industrial systems. Voluntary driver inputs at the wheel and pedals concentrate below about 5 Hz, with neuromuscular reflex and corrective content extending higher; the McRuer-style closed-loop crossover for routine driving sits much lower than that. The simulator bandwidth above the driver’s voluntary input range is needed for cueing fidelity rather than input recognition.
DIL vs MIL, SIL, HIL And VIL
The “in-the-loop” family is sequential, not competing. MIL (Model-in-the-Loop) runs the control algorithm against a plant model on a host PC, often slower than real time. SIL (Software-in-the-Loop) runs compiled code against the same plant. PIL (Processor-in-the-Loop) adds the target processor for timing-representative testing. HIL (Hardware-in-the-Loop) puts a real ECU on a bench against a real-time plant model with fault injection — the standard tool for ECU validation, OBD compliance and ISO 26262 functional safety work. VIL (Vehicle-in-the-Loop) is less standardised in industry usage than the earlier acronyms; in one common interpretation it places a complete vehicle on a chassis dyno or test rig with sensors fed simulated environment data, while other definitions include real test-track operation against virtual traffic. Ansible Motion and IAAPS launched one such facility in 2025. DIL slots into this chain at the point where the human driver becomes the test subject — typically for subjective tuning (steering feel, ride, ADAS warning timing) or any behaviour where the human is part of the safety case (L2/L3 handover, mode confusion).
Motion Platforms And Cueing
The dominant motion platform architecture is the six-degree-of-freedom Stewart platform (hexapod) — six prismatic actuators connecting a fixed base to the moving cockpit, providing surge, sway, heave, roll, pitch and yaw. Mid-tier OEM rigs and most motorsport simulators use a hexapod. Above that, large-envelope rigs add long-stroke linear rails to extend perceived acceleration cues and yaw range beyond the limits of a classical hexapod. AB Dynamics’ Ansible Motion Delta S3 puts the hexapod on 4–10 m linear rails and carries a 500 kg cockpit. VI-grade’s DiM250 and DiM400, and Dynisma’s DMG-1 and DMG-360XY, sit in the same large-envelope tier with multi-metre translation envelopes and dedicated motion-cueing strategies. Vendors quote peak actuator capability and translation distance rather than vehicle-equivalent sustained accelerations, which the cockpit cannot literally reproduce.
The motion-cueing algorithm (MCA) decides what fraction of the chassis acceleration gets rendered, how, and when to recentre the platform. Classical washout filters — high-pass on translation, low-pass on tilt — are simple to tune and robust, but use the workspace inefficiently. Adaptive and optimal cueing schemes vary gains with scenario or driver state. Model Predictive Control (MPC) approaches bake the platform’s actuator limits into the cueing problem and are increasingly common on premium rigs. Human vestibular models (Young–Meiry, Telban–Cardullo) are used to weight what the driver will actually perceive, so the platform spends its limited envelope on cues the driver can detect.
Force feedback through the steering wheel is the single most important tactile channel. Engineers detect torque irregularities, on-centre dead-band, non-linear gradients and parasitic friction at very small amplitudes. The steering loading unit and the EPS model behind it often get more attention than the motion platform itself when tuning vehicle dynamics, particularly for steering-feel work.
Vehicle Models, Scene Engines And The Software Stack
A DIL stack is usually built from three independently-sourced layers. The first is the vehicle dynamics model — IPG CarMaker / CarRealTime, VI-grade’s VI-CarRealTime, Mechanical Simulation’s CarSim/TruckSim, AVL VSM, or a customer’s in-house multi-body model wrapped through FMI or a Simulink interface. AVL and Ansible Motion announced a toolchain integration in March 2026 aimed at simplifying this interface between real-time vehicle model and motion platform. Tyre models in particular set fidelity: MF-Tyre (Magic Formula) is the steady-state handling default, MF-Swift adds rigid-ring dynamics, FTire supports substantially higher-frequency road-response modelling than rigid-ring tyre models and handles short-wave road obstacles, with the usable bandwidth depending on solver step size and road input, and CDTire/Realtime is the semi-empirical derivative used for ride and harshness work.
The second layer is the driving environment, sometimes called the scene engine. rFpro is the specialist incumbent and is widely used across Formula 1 and by a growing list of OEMs and tier-one ADAS developers, although team-by-team toolchains are not publicly disclosed. Competitors include AVSimulation SCANeR (which sits inside Ansys VRXPERIENCE), dSPACE AURELION, Cognata, NVIDIA DRIVE Sim, ESI Pro-SiVIC and Mathworks RoadRunner. The environment is what the driver sees, what the sensors see, and where the traffic actors live. ASAM OpenSCENARIO and OpenDRIVE are the industry-standard interchange formats for scenes and scenarios; the wider scenario-based validation ecosystem that has grown around UNECE R157 commonly uses them, although the regulation itself does not mandate either format.
The third layer is the real-time hardware. Speedgoat (Simulink Real-Time targets), dSPACE SCALEXIO and National Instruments real-time platforms are the common hosts. The same hardware often runs HIL and DIL workloads in different configurations.
Fidelity Classes — What Separates A Desktop Rig From An F1 Simulator
There is no industry-standard taxonomy for DIL fidelity, but the practical bands are: desktop sim-racing tier (wheel, pedals, single monitor) used for familiarisation and basic ADAS scenario work; fixed-base engineering tier with a real cockpit and force-feedback wheel for HMI and steady-state vehicle dynamics; 3-DOF motion (pitch, roll, heave) for HMI and introductory vehicle dynamics; 6-DOF hexapod for mainstream OEM ride-and-handling and ADAS work; and 9-DOF large-envelope rigs with translation rails for premium OEM and Formula 1.
The engineering questions that separate the tiers are motion-subsystem bandwidth (vendors quote figures approaching 100 Hz for small-amplitude motion responses on the leading platforms), end-to-end transport delay (low tens of milliseconds on the leading rigs), tyre-model bandwidth, scene fidelity for physically-based sensor rendering, and steering torque rendering bandwidth and resolution. F1 teams typically run 150–200 laps on a Thursday and around 12 hours of simulator time on a race-weekend Friday — usage levels that make latency, motion bandwidth and uptime first-order procurement concerns.
Major DIL Simulator Vendors
The market has consolidated around half a dozen serious players. rFpro is a dominant scene-engine specialist, widely used across Formula 1 teams and by OEMs and tier-one suppliers for sensor-realistic environments. Its digital twins of the Hakone Turnpike, Tokyo Shuto Expressway, Tōmei Expressway, a rural Warwickshire road and a 36 km Los Angeles loop now serve as standard development environments. AB Dynamics, which acquired Ansible Motion in 2022, sells the Delta series — S1, S2, the S3 large-envelope translation rig, and the new Delta T1 Sport compact motorsport simulator launched in May 2026. VI-grade, owned by Spectris since 2018, builds the DiM250 and the larger cable-driven DiM400.
Dynisma, the Bristol-based firm founded by Ash Warne (ex-Ferrari, ex-McLaren F1 simulator lead, now CTO), sells the DMG-1 and DMG-360XY and has publicly named Ferrari and McLaren Automotive among its customers. The company quotes motion-subsystem latency below 5 ms (measured at the platform IMU) and motion bandwidth above 100 Hz on its top rigs; end-to-end performance depends on the rest of the install. Cruden, a Dutch firm, builds the Hexatech 6-DOF platform and the Panthera software stack and has a long-standing motorsport and OEM customer base. dSPACE supplies the SCALEXIO real-time hardware, ASM vehicle dynamics models and AURELION sensor sim used by many integrators rather than a complete turn-key simulator. IPG Automotive’s CarMaker is more typically used MIL/HIL but is regularly integrated into DIL rigs through partners.
ADAS And Autonomous Vehicle Validation
Most of the recent investment in DIL has been driven by ADAS and autonomous-vehicle development. ISO 21448 (SOTIF) and UNECE R157 (the EU’s Automated Lane Keeping System regulation) both incorporate scenario-based assessment, and a credible safety case increasingly draws on an audit trail running from MIL through HIL, DIL and VIL. NHTSA’s proposed AV STEP framework, published in the Federal Register in January 2025, is moving in the same direction; Congress has appropriated $3.5 million in FY2023 and $4 million in FY2024 specifically for NHTSA’s virtual review and validation work on automated vehicles.
Camera, lidar and radar sensor models inside the DIL — physically-based rendering for cameras, ray-traced or neural-field models for lidar, Doppler-correct radar — let manufacturers run millions of scenario variants that would be prohibitively expensive on the road. KTM’s adoption of rFpro for headlight development in early 2026 is one example of how this is propagating beyond passenger cars. DIL also remains the only practical tool for human-factors testing on L2/L3 handover, mode confusion and HMI clarity, since those questions cannot be answered without a real driver.
Where DIL Simulators Are Going
Four trends matter. First, regulatory acceptance of simulator evidence in homologation submissions — UNECE R157 already incorporates scenario-based assessment as part of the ALKS safety case, and the NHTSA AV STEP framework is moving toward virtual review and validation. Second, sensor-model fidelity is moving from geometric to phenomenological, with physically-based rendering and neural-field methods for lidar and radar. Third, cloud-distributed simulation lets OEMs sweep millions of scenario variants in parallel. Fourth, digital twins of real roads are moving from one-off marketing pieces to a library that simulator vendors can rent. Many OEMs are increasingly positioning DIL as a formal verification stage rather than purely a late-stage subjective-development tool — the human-perceived attributes of a passenger car (steering feel, ride, HMI handover) are exactly the things that cannot be signed off any other way, even if SIL, HIL and cloud-scale scenario simulation remain the dominant pillars of ADAS and AV validation overall.
Bottom Line For Engineers Choosing A DIL Strategy
If your work is vehicle dynamics tuning, ADAS handover, HMI development or AV scenario validation, you need a DIL programme — the only question is what tier of facility, and whether to own, lease or rent simulator time. For top-tier motorsport, the answer is unambiguous: leading programmes operate dedicated simulator facilities, although smaller constructors, junior formulas and customer racing programmes often rent simulator access rather than owning rigs. For passenger-car OEMs, a phased approach — fixed-base for HMI, 6-DOF for development, 9-DOF or large-envelope for sign-off — is the norm. For tier-one suppliers and ADAS start-ups, time on a shared facility is increasingly the way in.
For more DIL simulator news, click here.
Frequently Asked Questions
A DIL simulator is a Driver-in-the-Loop simulator: a real-time vehicle development tool in which a human driver controls a virtual car through a real cockpit with force-feedback steering and pedals, a motion platform that renders chassis accelerations, and visual and audio cues. Most of the vehicle and environment are simulated, although some DIL rigs incorporate real ECUs, steering hardware or brake-by-wire subsystems in a HIL-style integration. The rig must respond to the driver and feed back cues at rates the human nervous system reads as continuous.
HIL (Hardware-in-the-Loop) places a real ECU on a bench against a real-time vehicle plant model and is used for ECU validation, fault injection and functional safety work. DIL puts a real human driver in the loop instead of (or alongside) the ECU. HIL is the standard tool for software and ECU correctness; DIL is the tool for any question where the human is part of the answer, such as steering feel, ride quality, ADAS warning timing or L3 handover behaviour.
End-to-end transport delay, from driver input to perceived motion or visual response, is the single most important system-level metric on a DIL rig. Drivers detect very small temporal inconsistencies through the steering wheel and the vestibular system and form invalid subjective judgements if the delay is too high. Best-in-class motion and control subsystems achieve sub-10 ms latencies; end-to-end perceived latency on the leading rigs typically sits in the low tens of milliseconds, with older industrial rigs running substantially higher. As a rule of thumb, the closer end-to-end delay gets to single-digit milliseconds the more faithful the rig feels.
rFpro is a scene-engine and driving simulation environment software vendor; it does not build motion platforms. AB Dynamics, through its Ansible Motion subsidiary, builds large-envelope motion platforms (the Delta series) and turn-key DIL rigs. Dynisma is a UK motion-platform specialist founded by an ex-Ferrari F1 simulator lead and highlights industry-low subsystem latency and high motion bandwidth on the DMG-1 and DMG-360XY, with end-to-end performance depending on the rest of the installation. They are often combined: an OEM might use an Ansible Motion or Dynisma platform with rFpro running the scene.
Increasingly, yes. UNECE R157 (the EU ALKS regulation) already incorporates scenario-based assessment as part of the safety case, and the simulation evidence behind it typically passes through a DIL programme for human-factor work. NHTSA’s proposed AV STEP framework is moving in the same direction. The longer-term picture is that a credible AV homologation file will rely on an audit trail running from MIL through SIL, HIL, DIL and VIL into limited on-road validation, not on millions of test miles alone.
