Automated and connected driving is one core element of future mobility, which must be safe, environmentally friendly, intelligent and sustainable. A prerequisite for autonomous driving functions is functional safety based on precise real-time detection of the dynamic traffic environment. Radar sensors play a key role here, as they operate reliably regardless of lighting conditions or weather, unlike optical sensors such as cameras or LiDAR. Before being used in road traffic, their functional safety must be assured by reliable test methods. Traditionally, verification and validation were distance-based and involved hundreds of millions of kilometers of driving. These tests, while realistic, have become prohibitively expensive and time-consuming. Instead, scenario-based testing is used and is under standardisation nowadays, which optimally combines real hardware with virtual driving scenarios. For the scenario-based testing of the installed performance of automotive radar, a powerful method has been developed that enables realistic, technology-agnostic, reproducible and efficient testing of radar sensors in a virtual electromagnetic environment. The over-the-air vehicle-in-the-loop (OTA/ViL) approach integrates the sensor system under test, a radar target simulator and a propagation model into a closed simulation test loop. This enables automated testing under realistic conditions and closes the gap between field tests and software-in-the-loop simulations. Proven through publicly funded projects and already adopted by international car manufacturers, this method exemplifies a very successful technology transfer from science to economy that paves the way for the mobility of the future.

INTRODUCTION

Automated and connected driving must be safe, efficient and reliable, and that begins with perception. Automotive radar is central to this task: unlike cameras and LiDAR, it operates robustly regardless of weather or lighting conditions, making it indispensable to a fused sensor suite and safety-critical advanced driver assistance systems (ADAS).^1,2 Before such sensors reach the road, their installed performance must be verified by trustworthy test methods.

Traditionally, that verification relied on distance-based road testing, approaching billions of kilometers under varying conditions.³ Field testing of this kind is poorly reproducible and, as development cycles shorten and system complexity grows, increasingly impractical — driving demand for reproducible, scalable and efficient alternatives.⁴

Virtual test environments address this need by integrating real radar hardware into closed loop frameworks that combine virtual driving scenarios with controlled electromagnetic conditions.⁴ The OTA/ViL approach is one such method: it emulates realistic, dynamic targets typical for a road traffic environment through electronically generated radar signals, combining radar hardware, wave-propagation models and a scenario simulator into a fully integrated closed loop.^5,6 This enables reproducible scenario testing of radar perception in the installed state within a controlled laboratory and thus contributes to a holistic safety assurance. Developed as part of the virtual road simulation and test area (VISTA) of the Thuringian Center of Innovation in Mobility (ThIMo) at Technische Universität Ilmenau (TU Ilmenau) (see Figure 1) and matured through a series of publicly funded industrial-academic projects, the platform was transferred to the industry by dSPACE GmbH and adopted by international OEMs for system development and end-of-line testing. This article summarizes its technical foundation, application areas and industrial impact.

SYSTEM OVERVIEW

Architecture of the OTA/ViL Test Setup

Figure 2 shows the system architecture. At its core, a hardware-in-the-loop (HiL) simulator coordinates all computational and physical processes in real time, modelling dynamic traffic scenarios and converting them into radar-relevant parameters of all targets under consideration, i.e., range, speed, radar cross-section (RCS)and angular direction — at a 1 ms update rate.⁷ The main constituents of the VISTA OTA/ViL setup are:

• Radar-under-test (RUT) and data interface: A fully integrated radar mounted in its operational position behind the front grill and radome, be it a commercially available system or a research radar. Raw data access enables in-depth evaluation beyond the object list normally exposed to the electronic control unit, while protecting the sensor’s proprietary signal processing.
• Radar target simulators (RTS): Generate digital radar echoes with precise range (5.5 to 1000 m), relative velocity (±700 km/h) and RCS at a 1 kHz update rate; MIMO air interfaces allow multiple simultaneous targets.⁸
• Antenna positioners and air interface: Seven independent three-axis masts on a dual-rail system steer, receive and transmit antenna pairs across azimuth (ϕ = ±50), elevation (è = -11 degrees/ +15 degrees) and pitch (±30 degrees) with < 0.5 degrees angular resolution; the modular design is easily expanded or reconfigured.⁹
• HiL framework and software: A dSPACE Scalexio system synchronises target simulation, positioner actuation and RUT stimulus over CAN and Ethernet; scenarios are defined in the automotive simulation model environment in Simulink.
• Calibration, collision avoidance and handover: Mandatory pre-simulation calibration of the mast motors avoids position errors, while a state machine governs collision-free operation, real-time target-to-mast assignment, pre-positioning for minimal latency and handover when masts reach their movement limits.⁹

Working Principle

Each test begins by configuring a virtual traffic scenario in the HiL environment. The RUT emits radar signals that the RTS intercepts, modifies according to the scenario’s target parameters and retransmits in real time, while the antenna positioners steer to the corresponding angles of arrival. The RUT processes these synthetic returns and feeds its perception output back to the HiL to close the loop, with all data logged for validation.⁷

TESTING PROCEDURE AND METHODOLOGY

The OTA/ViL workflow forms a closed loop that links a virtual traffic scene with real radar sensors in their installed state. A scenario is defined on the HiL host computer, translated into radar-relevant parameters (range, relative velocity/Doppler shift, RCS, azimuth, elevation) and transferred to the radar target simulator and the antenna positioners. The RUT emits and shortly later receives the synthetically “backscattered” signals over the air and returns its object list to the HiL for logging and evaluation — closing the test loop. This end-to-end chain is repeatable, time-coherent and robust, yet preserves installation effects that pure software simulation misses. A key feature is multi-level observability. Data are captured at three abstraction layers (AL):

• AL1 (model): Radar parameters associated with the model scenario simulated in the HiL
• AL2 (stimulation): Signals commanded to RTS and positioners over the air, stimulating the RUT
• AL3 (sensor): Objects detected by the installed RUT.

By comparing all three abstraction layers at millisecond granularity, the method reveals where errors originate, in simulation, radar signal emulation or sensor/installation, to assess limitations and devise appropriate countermeasures. The result is on-demand, repeatable ADAS testing (e.g., maneuvers such as automated emergency braking (AEB), adaptive cruise control, blind spot warning and lane change) that complements proving-ground testing with laboratory-grade control, reduces costs and risks and preserves the relevant installed-state effects that determine real-world performance.

Radar Signal Vectors

Each simulated point target is represented by a time-dependent state vector⁵

S(t_i) = {R, v, σ, ϕ, θ},
i = 1,...,N

where R is range, v the relative speed derived from the Doppler shift, σ the RCS and (ϕ, θ) the azimuth and elevation angles. At runtime, the RTS electronically synthesises R, v and σ, while the antenna positioners realise ϕ and θ through controlled motion. This division of tasks preserves the physical angles of arrival while enabling precise control over kinematics and target strength. Following the abstraction layers introduced above, inter-layer deltas are analysed to validate the achievable fidelity (e.g., azimuth tracking, RCS deviation) and to tune controllers and absorber layouts.

Scenario Generation and Automation

Scenarios are curated into a library covering, e.g., longitudinal highway, urban crossing and 3D (elevation) cases with static and dynamic objects and typical manoeuvres such as car-following, braking to standstill, cut-in/cut-out, on-ramp merges and occlusions. For each participant, trajectories, speeds and separations are parametrised, and RCS profiles are assigned from lookup tables fed through measurements in VISTA. The compiled inputs are streamed to the RTS and positioners for real-time execution: the positioners discretise the continuous angles of arrival into high-rate azimuth, elevation and pitch setpoints, with a mast-to-target assignment and handover strategy that enable multi-object dynamics mirroring real traffic. The framework configurability supports scenarios derived from standards such as Euro NCAP^10,11 as well as use cases anticipated for higher automation levels.

The HiL scheduler enforces synchronicity: Scenario parameters update every 1 ms, while typical RUT object list updates occur at a much slower repetition rate of 50 ms. All streams are time-stamped; intermediate samples are interpolated to a 1 ms grid for correlation across all three abstraction layers. This tight synchronisation is crucial for evaluating controller stability (e.g., positioner inertia or overshoot) and for attributing perception lag to sensor versus testbed.

Before real-time execution, the masts are calibrated and referenced against their physical motion limits. At the same time, a state machine governs collision-free operation, real-time target-to-mast assignment, pre-positioning and handover between adjacent masts to maintain uninterrupted tracking.⁹

Testing Procedure

• Data logging and post-processing: All relevant streams — raw radar signals, the RUT object list, HiL control commands and mast actuator positions — are logged simultaneously with synchronised timestamps across Ethernet, CAN and wireless interfaces. Post-processing aligns these heterogeneous datasets onto a unified timeline, removes noise and incomplete samples and converts them into standardised formats for analysis.
• Data analysis: The synchronised data are compared across the three abstraction layers; deviations reveal systematic errors such as latency, angular offsets or signal attenuation. Metrics such as detection probability, range, Doppler accuracy and angular resolution are extracted to evaluate sensor performance. Overlays of simulated trajectories with sensor detections support intuitive assessment of perception quality.
• Verification with reference data: Results are benchmarked against independent reference measurements from proving grounds — decimetre-accurate position and velocity from high-resolution radars and differential navigation, plus vehicle-dynamics data from real-time kinematic kits.¹² Comparing the synthetic signal vectors against these references quantifies quality indicators such as range accuracy, Doppler precision and angular tracking reliability, confirming that laboratory results correlate with real-world sensor behavior.

VALIDATION AND BENCHMARKING

Validation within the OTA/ViL framework combines standardised Euro NCAP scenarios¹⁰ with dedicated, research-driven tests that probe sensor performance more directly.¹³

AEB car-to-car protocols, such as the rear-stationary and cut-across cases, provide a reproducible baseline for cross-system benchmarking, while structured motion patterns — figure-of-eight, slalom and circular driving — stress target reidentification, rapid lateral dynamics and angular/Doppler consistency under sustained curvature. The setup emulates these by converting the exact protocol parameters (vehicle speed, overlap, braking profiles, lateral tolerances) into real-time radar signal vectors,¹⁴ so the radar perceives the emulated targets as in real traffic. Environmental effects such as noise, clutter and road-surface properties can also be emulated to enhance realism further.

Example Scenario Depicting the OTA/ViL Functional Performance

This section provides an overview of a typical crossing junction scenario, as shown in Figure 3a. This scenario is intrinsically challenging because the transverse motion is hard to detect by a radar due to the very small Doppler shifts along the projected trajectory, see Figure 3b. A real research vehicle equipped with a commercial automotive radar¹⁵ mounted in the front grill served as the ego vehicle, see Figure 3c, and kept stationary in the test. The target vehicle was modelled as a virtual mid-sized car, with its RCS characteristics prerecorded and represented by a nearest-possible point source.

In the crossing scenario, the target vehicle follows a predefined straight path perpendicular to the ego vehicle. The separation between the ego vehicle and the junction was set to R = 70 m, ensuring that the target car would appear within the maximum sensing range of the RUT, which was 100 m in this case. The target car followed a straight path from right to left at a constant speed of 20 km/h (around 5.6 m/s) throughout the test, which continued until the entire trajectory was covered.

The radar signal vectors were recorded at all three abstraction levels (see Figure 4): AL1 (model) as blue curves, the AL2 OTA stimulus as black curves and the AL3 radar measurement as red crosses, the latter limited to the window in which the target is detected. The range (see Figure 4a) is parabolic, with its minimum where the target car is nearest. The relative velocity (see Figure 4d) changes dynamically due to trajectory-induced azimuthal variations. The azimuth (see Figure 4b) ideally spans ϕ = -45 to 45 degrees, but the target is detected only near ϕ = ±35 degrees owing to the limited field-of-view¹⁶ and the lower RCS at the car edges; the step around 12 to 15 seconds stems from a geometrical jump of the modelled nearest point representing the target. The RCS (see Figure 4c), reconstructed at AL2 from the AL1 model and measured at AL3, agrees well across all layers. Overall, range, velocity, azimuth and RCS show very good agreement across abstraction levels, with the remaining discrepancies presently under investigation. These differences can generally be attributed to system limitations, paving the way for further improvements in test installations.

Optimization Potential of the OTA/ViL Approach

While OTA/ViL closes the gap between numerical simulation and real-world drive tests, several constraints shape its performance. Although the RTS can digitally create targets as close as 0.1 m, cable and processing delays and the antenna RUT separation raise the effective minimum distance to R ≈ 13 m⁹; this lies within the scope of most forward-looking radars but restricts very close-range interactions. The positioner mechanics bind dynamic performance: for safety and energy efficiency, the carbon-fibre masts are limited to 5 m/s², so high speed lateral crossings can introduce angular tracking errors that grow with target velocity. These errors can reach several degrees at 90 km/h and 20 m range.⁹ Anticipatory control that pre-compensates the positioner trajectory from predicted motion offers one path to mitigate this. The integrated RTS generates up to four point scatterers per unit — sufficient for basic vehicles but not for traffic situations — which also stresses the mast assignment algorithms, where advanced point-cloud synthesis or hybrid mechanical–electronic beam steering could help. Eventually, mast-positioning errors at short range magnify into angular offsets at longer range, so higher-fidelity calibration is advisable. Addressing these through anticipatory motion planning, improved calibration and richer target models would broaden the range of scenarios that OTA/ViL can capture while preserving its reproducibility and efficiency.

SUCCESS STORY: TRANSFER FROM ACADEMIA TO COMMERCIAL TEST SOLUTIONS

Exemplary Research-to-Industry Pipeline

The presented OTA/ViL approach originated from academic research at ThIMo at the Technische Universität Ilmenau, Germany. First established in the publicly funded research and development project, SafeMove “installed performance evaluation of automotive radar systems over-the-air,”⁵ the system was conceived as a modular, reproducible method for testing radar sensors in their installed state within a virtual electromagnetic environment. Its scientific novelty lay in combining vehicle-in-the-loop simulation with over-the-air stimulation, enabling realistic yet fully controlled validation of advanced driver assistance functions.

The successful proof-of-concept paved the way for a technology transfer to dSPACE GmbH, which industrialized the setup and launched it under the brand name VERIDRIVE.¹⁷ From this base, the solution has been delivered globally, with more than a dozen OTA/ViL systems deployed over the past five years. A milestone was reached with its adoption by the FAW Group (Hongqi brand) in China,¹⁸ where the setup was integrated into production-level radar testing workflows. Figure 5 shows an example of an OTA/ViL system in action at the FAW Group site. This impressive transfer chain illustrates how scientific innovation at university level can mature into a commercial product with global reach.

Integration into Industry-Grade ADAS Testing Workflows

In industrial practice, VERIDRIVE has been incorporated into ADAS development and production validation chains. At FAW, the system supports the development of highway pilot functions, where accurate and repeatable radar validation is critical for longitudinal and lateral control. By providing realistic emulation of Euro NCAP test cases and complex traffic scenarios in a closed laboratory loop, the system allows manufacturers to evaluate perception performance long before extensive proving-ground mileage is accumulated.

Beyond OEM-focused workflows, the OTA/ViL test method has also been applied to the periodic technical inspection (PTI) of ADAS functions, addressing a regulatory gap in which safety-critical systems were previously not checked after vehicle delivery. Since 2020, dSPACE has worked with the Korea Transportation Safety Authority (KOTSA) on PTI-related projects, and in Germany the KÜS DRIVE center has operated a VERIDRIVE-based solution since 2021.¹⁹ In parallel, the CITA ADAS taskforce is preparing standards for PTI testing of driver assistance systems, with OTA/ViL strongly represented. Equipment suppliers such as Maha have already started developing workshop test stands (e.g., MAST) with VERIDRIVE as a key component.

Industry Feedback and Benefits

Feedback from early adopters, most notably FAW Hongqi, highlights the tangible benefits of the OTA/ViL approach. As Qu Jindai (FAW Hongqi spokesperson) notes, ”dSPACE has provided our development department with a comprehensive solution for vehicle-in-the-loop (VIL) testing, which is successfully used for dynamic, scenario-based testing of automated real vehicles.”

Building on this experience, three advantages recur across the deployment phase. Firstly, development cycles are shortened: shifting a substantial portion of validation into a virtual laboratory environment markedly reduces proving-ground and field operational testing. Secondly, safety-critical radar performance is validated earlier: highly reproducible lab conditions expose weaknesses and corner-case failures at a stage where design changes are still cost-effective. Thirdly, regulatory readiness improves: by reproducing standardised Euro NCAP scenarios and aligning with emerging PTI discussions (e.g., amendments to Directive 2014/45/EU), the same setup supports both current compliance and future inspection regimes.

Taken together, the development of the OTA/ViL testbed is a textbook case of knowledge transfer from academia to industry. Originating from ThIMo, industrialized and scaled by dSPACE and adopted by OEMs, the technology now serves a dual mission: accelerating ADAS development today while positioning manufacturers for tomorrow’s regulatory requirements.

CONCLUSION AND FUTURE WORK

The OTA/ViL method combines ViL simulation with OTA stimulation to test automotive radar in its installed state, closing the gap between purely virtual simulation and resource-intensive field testing. It enables early, reproducible detection of sensor weaknesses under laboratory conditions, reduces proving-ground mileage and aligns with emerging Euro NCAP and PTI requirements. Its path from an academic concept to a commercial product and its adoption by OEMs illustrate how research-driven innovation can reach global industrial use.

Future work will further improve the fidelity of laboratory emulation by integrating clutter, noise, multipath and extended multi-scatterer targets. It will extend OTA/ViL principles to multi-sensor fusion with lidar, cameras and V2X. Harmonised and standardised OTA/ViL frameworks will be key to comparability across laboratories and to supporting certification. With continued development, the method is poised to become a cornerstone of efficient, reliable sensor validation for automated and connected mobility.

ACKNOWLEDGMENTS

This work was funded by the Federal Ministry for Education and Research (BMBF) under the projects acronym SafeMove (grant numbers 16ES0547K and 16ES0551), VIVID (grant numbers 16ME0172 and 16ME0164K), Deutsche Forschungsgemeinschaft (DFG) under project acronym 4CAD (grant number 503852364) and German Federal Ministry for Economic Affairs and Energy (BMWE) under the acronym CONTROL (grant 19DNS2501S). The authors gratefully acknowledge the support and valuable contributions of current and former colleagues, namely Felix Kreutz, Johannes Nagel, Philip Aust, Tobias Nowack, Christian Bornkessel and Thomas Dallmann.

References

1. H. Winner, S. Hakuli, F. Lotz and C. Singer, Eds., Handbook of Driver Assistance Systems: Basic Information, Components and Systems for Active Safety and Comfort. Cham, Switzerland: Springer, 2016.
2. Y. Zhang, Y. Liu and J. Wang, “Perception and sensing for autonomous vehicles under adverse weather: A survey,” ISPRS Journal of Photogrammetry and Remote Sensing, Vol. 194, 2023, pp. 40–68.
3. N. Kalra and S. M. Paddock, “Driving to safety: how many miles of driving would it take to demonstrate autonomous vehicle reliability?,” Transportation Research Part A: Policy and Practice, Vol. 94, 2016, pp. 182–193.
4. H. Winner, K. C. J. Dietmayer, L. Eckstein, M. Jipp, M. Maurer and C. Stiller, Eds., Handbook of Assisted and Automated Driving. Wiesbaden, Germany: Springer, 2023. doi: 10.1007/978-3-658-45276-6.
5. S. Buddappagari et al., “Over-the-Air Vehicle-in-the-Loop Test System for Installed-Performance Evaluation of Automotive Radar Systems in a Virtual Environment,” 2020 17th European Radar Conference (EuRAD), Utrecht, Netherlands, 2021, pp. 278–281, doi: 10.1109/EuRAD48048.2021.00078.
6. P. Aust et al., “VIVID – German-Japan joint virtual validation methodology for intelligent driving systems,” White Paper, 2023.
7. dSpace GmbH, “HiL Scalexio,” 2017, Web: https://www.dspace.com/en/pub/home/products/hw/simulator_hardware/scalexio.cfm.
8.  dSpace GmbH, “DARTS 9030-M,” Web: https://www.dspace.com/de/gmb/home/products/hw/darts/darts-9030-m.cfm.
9. M. E. Asghar, “Measurement and installed performance evaluation of automotive and locomotive antennas and radar systems testing in a virtual electromagnetic environment,” Ph.D. dissertation, Technische Universität Ilmenau, Ilmenau, Germany, 2024. doi: 10.22032/dbt.64608.
10. 10. European New Car Assessment Programme (Euro NCAP), “Test Protocol AEB Car-to-Car systems,” v3.0.1, Brussels, 2019. Web: https://cdn.euroncap.com/cars/assets/euro_ncap_aeb_c2c_test_protocol_v431_532926aad1.pdf.
11. M. Wisch, H. Fagerlind, L. Sulzberger and M. McCarthy, “Specifications for scenario definitions,” ASSESS Project, Deliverable D1.2, 2010, pp. 1–91.
12. S. B. Jayapal Gowdu, P. Aust, A. Schwind, F. Hau and M. A. Hein, ”Evaluation of scenario-based automotive radar testing in virtual environment using real driving data,” 2022 IEEE 25th International Conference on Intelligent Transportation Systems (ITSC), Macau, China, 2022, pp. 2379–2384, doi: 10.1109/ITSC55140.2022.9922366.
13. PEGASUS Method, PEGASUS Project, Web: https://www.pegasusprojekt.de/en/pegasus-method.
14. M. L. Nazar, M. Pourjafarian and M. A. Hein, ”Over-the-Air Virtual Scenario Emulation for Vehicle-in-the-Loop Testing with Improved Accuracy,” 22nd European Radar Conference (EuRAD), Utrecht, Netherlands, 2025, pp. 51–54, doi: 10.23919/EuRAD65285.2025.11233896.
15. Continental AG, “ARS510,” Web: https://www.continental-automotive.com.
16. Muller, “Survey on Ranging Sensors and Cooperative Techniques for Relative Positioning of Vehicles,” Sensors, Vol. 17, No. 2, p. 271, Jan. 2017.
17. dSPACE GmbH, ”Vehicle-in-the-Loop Testing,” dSPACE GmbH, n.d. Web: https://www.dspace.com/en/pub/home/applicationfields/foo/vehicle-in-the-loop-testing.cfm.
18. “FAW’s Hongqi: Is It Okay to Test ADAS/AD Functions with a Platform HIL System?,” dSPACE GmbH, Web: https://www.dspace.com/en/pub/home/applicationfields/stories/faw-hongqi-is-it-okay-phvil.cfm.
19. ”Korean Transportation Safety Authority Develops Test Methods for Driver Assistance Systems with Dürr and dSPACE,” dSPACE GmbH, April 20, 2023, Web: https://www.dspace.com/en/pub/home/news/dspace_pressroom/press/korean-transportation-safety-a.cfm.