AuRaSim: A Ray-Tracing Based Automotive FMCW Radar Simulator for Dynamic Scenarios

Zhaoze Wang; Changxu Zhang; Tai Fei; Christopher Grimm; Yi Jin; Claas Tebruegge; Markus Gardill

cs.ROOct 27, 2026

AuRaSim: A Ray-Tracing Based Automotive FMCW Radar Simulator for Dynamic Scenarios

Authors: Zhaoze Wang, Changxu Zhang, Tai Fei, Christopher Grimm, Yi Jin, Claas Tebruegge, Markus Gardill

Organizations: HELLA GmbH & Co. KGaA, 59552 Lippstadt, Germany · Department of Electronic Systems and Sensors, Brandenburg University of Technology, 03046 Cottbus, Germany · Department of Information Technology, Dortmund University of Applied Sciences and Arts, 44139 Dortmund, Germany

Abstract

Modern simulators can generate synthetic camera and LiDAR data with accurate annotations for downstream tasks. In contrast, automotive radar simulation remains less mature because it requires not only a 3D engine to model dynamic objects in the scenes but also electromagnetic material properties, antenna configurations, wave propagation mechanisms, and coherent signal synthesis. Existing frameworks are either proprietary or overly simplified, limiting the scalability of radar raw measurements. This paper presents AuRaSim, a ray-tracing-based automotive frequency-modulated continuous wave (FMCW) radar simulation framework for dynamic scenarios. Rather than implementing all components from scratch, AuRaSim builds on Blender for efficient dynamic-scene construction and Sionna RT for graphics processing unit (GPU) accelerated electromagnetic ray tracing, while introducing radar-specific adaptations for path-resolved propagation with complex coefficients, motion-aware Doppler modeling, and subsequently coherent multi-channel analog-to-digital converter (ADC) cube synthesis based on waveform configurations. Quantitative comparison between simulated and measured range-Doppler maps in digital-twin scenarios achieves a root-mean-square error (RMSE) of 0.147 and a peak signal-to-noise ratio (PSNR) of 16.93 dB, demonstrating the accuracy and fidelity of the synthetic signal. AuRaSim therefore provides a complete and extensible platform for reproducible automotive radar simulation and research. Project homepage: https://wangzhaoze.github.io/aurasim.

Figures & tables

Fig. 1: Principal propagation mechanisms considered by the Path Solver: specular reflection, diffuse scattering, thin-slab transmission, and edge diffraction.

Fig. 2: System overview of AuRaSim. A trajectory-driven dynamic digital twin and a configurable radar sensor are synchronized at each radar timestamp. The Path Solver produces an explicit path-level representation, which is subsequently converted into waveform-consistent multi-channel ADC samples and derived radar products.

Fig. 3: The Path Solver pipeline of AuRaSim. Black components correspond to the Sionna RT $\mathtt{PathSolver}$ , while red components indicate the radar-oriented adaptations introduced in AuRaSim. AuRaSim restricts the initial ray distribution to the radar field of view, applies path filtering before field evaluation, provides local hit-point velocities for Doppler calculation, and exports an explicit path-level representation for subsequent radar signal synthesis.

Fig. 4: Illustration of ray sampling. Uniform angular sampling exhibits nonuniform point density over the sphere, while spherical Fibonacci sampling provides a more uniform full-sphere distribution. The proposed FOV-aware Fibonacci sampling concentrates the same number of rays directly within the radar field of view.

Parameter	Value	Parameter	Value
Radar frame rate	30 Hz	Carrier frequency $f_{c}$	77 GHz
Tx/Rx antennas	$2\,\mathrm{TX}\times 4\,\mathrm{RX}$	MIMO scheme	TDM
ADC samples/chirp	128	Chirps per frame	510
ADC sampling rate $f_{s}$	4 MSps	Effective bandwidth $B$	0.67 GHz
Frequency slope $\mu$	$21\,\mathrm{MHz}/\mu\mathrm{s}$	Chirp duration	$\approx 60\,\mu\mathrm{s}$

TABLE I: Radar configuration used for both the real measurements and AuRaSim.

Fig. 5: Representative qualitative comparison of real and simulated data. From left to right: real camera image, rendered digital-twin view, RadarSimPy RD map, AuRaSim RD map, and measured RD map. We show two representative frames for each sequence. For all RD maps, the horizontal axis corresponds to radial velocity, with negative velocities on the left (moving away) and positive velocities on the right (approaching), while the vertical axis corresponds to range, with range increasing from bottom to top. The color intensity represents the normalized radar power in dB.

Fig. 6: Micro-Doppler analysis of a simulated walking pedestrian using AuRaSim. (a) Reconstructed walking motion and trajectories of selected body joints while the pedestrian moves toward the radar. (b) Simulated micro-Doppler spectrum generated from the dynamic scene. (c) Radial velocities of representative body joints. The periodic Doppler spread is consistent with the articulated limb motion of the pedestrian.

Scenario	Method	RMSE $\downarrow$	PSNR [dB] $\uparrow$
BMS1000	RadarSimPy [ 37 ]	0.298	11.94
BMS1000	AuRaSim	0.172	15.27
PMS2000	RadarSimPy [ 37 ]	0.189	15.46
PMS2000	AuRaSim	0.159	16.11
MLMS001	RadarSimPy [ 37 ]	0.233	12.90
MLMS001	AuRaSim	0.111	19.40

TABLE II: Quantitative comparison between measured and simulated range–Doppler maps in terms of RMSE and PSNR .

\downarrow

denotes lower is better and

\uparrow

denotes higher is better.

Exp.	Rays [ $\times 10^{6}$ ]	IV-D 1 FOV Ray Sampling	IV-D 2 Path Filtering	IV-D 3 Hit-Point Doppler	PSNR [dB] $\uparrow$	Runtime [s/frame] $\downarrow$
E1	0.1	$\checkmark$	$\checkmark$	$\checkmark$	16.81	0.672
E2	1	$\checkmark$	$\checkmark$	$\checkmark$	18.21	0.690
E3	5	$\checkmark$	$\checkmark$	$\checkmark$	18.71	0.774
E4	10	$\checkmark$	$\checkmark$	$\checkmark$	19.40	0.838
E5	10	$\times$	$\checkmark$	$\checkmark$	18.59	0.859
E6	10	$\checkmark$	$\times$	$\checkmark$	15.46	0.826

TABLE III: Ablation study on the MLMS001 scenario. PSNR measures the similarity between simulated and measured range–Doppler maps, while runtime reports the corresponding ray-tracing time per radar frame.

Fig. 7: Qualitative ablation results on the MLMS001 scenario. (a) RGB reference image. (b)–(e) Simulated range–Doppler maps using 0.1, 1, 5, and 10 million rays, respectively. (f) Result with 10 million rays without the proposed FOV-aware ray sampling. (g) Result without path filtering. (h) Result using the conventional object-level Doppler model instead of the proposed local hit-point velocity formulation. (i) Measured radar data. The red boxes highlight the range–Doppler region associated with the vehicle-wheel response and are used to visualize differences in the corresponding micro-Doppler structure.

Fig. 8: Qualitative comparison of synthetic-array and explicit-array MIMO modeling against measured range–azimuth (RA) maps. From top to bottom: PMS2000 , BMS1000 , and MLMS001 . From left to right: recorded camera image, Blender-rendered scene, synthetic-array result, explicit-array result, and measured RA map. The radial and angular coordinates indicate range and azimuth, respectively. Each RA map is independently normalized and displayed on a relative power scale from $-60$ to $0$ dB.

Scenario	MIMO Mode	PSNR [dB] $\uparrow$	Runtime [s/frame] $\downarrow$
BMS1000	Synthetic	16.43	0.52
BMS1000	Explicit	16.46	1.26
PMS2000	Synthetic	17.73	0.52
PMS2000	Explicit	17.76	1.28
MLMS001	Synthetic	18.51	0.83
MLMS001	Explicit	18.70	2.07

TABLE IV: Comparison of PSNR and runtime of MIMO array modeling modes with range–azimuth maps.

Physics-Based Simulation

Physics-based simulation models physical systems with equations and numerical solvers, increasingly augmented by learned models for speed and control.

See all 231 papers

Mar 11, 2025

Simulating Automotive Radar with Lidar and Camera Inputs

Low-cost millimeter automotive radar has received more and more attention due to its ability to handle adverse weather and lighting conditions in autonomous driving. However, the lack of quality datasets hinders research and development. We report a new method that is able to simulate 4D millimeter wave radar signals including pitch, yaw, range, and Doppler velocity along with radar signal strength (RSS) using camera image, light detection and ranging (lidar) point cloud, and ego-velocity. The method is based on two new neural networks: 1) DIS-Net, which estimates the spatial distribution and number of radar signals, and 2) RSS-Net, which predicts the RSS of the signal based on appearance and geometric information. We have implemented and tested our method using open datasets from 3 different models of commercial automotive radar. The experimental results show that our method can successfully generate high-fidelity radar signals. Moreover, we have trained a popular object detection neural network with data augmented by our synthesized radar. The network outperforms the counterpart trained only on raw radar data, a promising result to facilitate future radar-based research and development.

Peili Song, Dezhen Song, Yifan Yang +2

Institute of Robotics and Automatic Information System, Nankai University, Tianjin 300350, China; Tianjin Key Laboratory of Intelligent Robotics, Tianjin 300350, China; and also with TBI center, Nankai University, Tianjin 300350, China · Department of Robotics, Mohamed Bin Zayed University of Artificial Intelligence (MBZUAI), Abu Dhabi, United Arab Emirates

Synthetic Data Augmentation Millimeter-Wave Radar mmWave Radar Sensing Autonomous Driving Perception Synthetic Data Generation for Robotics Autonomous Driving Simulation

Ask PDF

May 25, 2026

RadarSim: Simulating Single-Chip Radar via Multimodal Neural Fields

Radars are an ideal complement to cameras: both are inexpensive, solid-state sensors, with cameras offering fine angular resolution, while radars provide metric depth and robustness under adverse weather. However, radar data is more difficult to interpret than camera images and varies significantly between sensors, necessitating increased reliance on simulation for prototyping sensors and processing pipelines. Recent work treating radar reconstruction as a novel view synthesis problem has shown great promise in reconstructing radar-relevant geometry and simulating low-level radar data. However, such methods are constrained by the low spatial resolution of the underlying radar. To address this, we propose a unified differentiable renderer, RadarSim, which leverages the high angular resolution of RGB cameras to generate Doppler radar range images from a camera-initialized neural field. Using a novel data set of calibrated radar camera recordings from a custom hand-held rig, we demonstrate that RadarSim produces sharper geometry and Doppler range frames than radar-only reconstructions.

Chuhan Chen, Tianshu Huang, Akarsh Prabhakara +5

Carnegie Mellon University · Bosch Research · University of Wisconsin–Madison

Camera-Radar Fusion 3D Reconstruction Neural Rendering

Ask PDF

Jun 24, 2026

RadarTwin: Scene-Specific mmWave Radar Simulation and Learning for Mobile Indoor Perception

Millimeter-wave (mmWave) radar perception is limited by data scarcity: models trained on existing radar datasets fail to generalize to new objects, environments, and sensing trajectories. We present RadarTwin, a framework for generating deployment-specific radar training data before real data collection. Given a 3D reconstruction of a target space (phone LiDAR, robot-mounted sensing, or RGB-to-3D), RadarTwin uses a vision-language model to infer radar-relevant surface materials and a physics-based ray tracer to synthesize raw frequency-modulated continuous-wave (FMCW) radar measurements with multi-bounce propagation. To study what transfers from simulation to reality, we collect a paired real-simulated dataset spanning household objects, material classes, distances, rotations, translations, and mobile sensing trajectories. We show that simulated and real radar share the same object-discriminative shape and material features, and that modeling the environment's multipath is essential to matching real measurements. A representation trained on simulation alone recognizes real objects at 2.5 times chance with no real radar labels, and a few labeled examples raise this to 95.3% on a 12-way recognition task. RadarTwin enables training radar perception for a new space before any real radar data is collected there.

Emily Bejerano, Federico Tondolo, Devang Gupta +5

Columbia University New York, NY, USA · University of California, Merced Merced, CA, USA

Sim-to-Real Transfer Millimeter-Wave Radar mmWave Radar Sensing Synthetic Data Generation for Robotics

Ask PDF

Abstract

Physics-Based Simulation

Explore similar work

Simulating Automotive Radar with Lidar and Camera Inputs

RadarSim: Simulating Single-Chip Radar via Multimodal Neural Fields

RadarTwin: Scene-Specific mmWave Radar Simulation and Learning for Mobile Indoor Perception