DRaGAN: Dynamic Graph Attention Network for Multi-Radar Point Cloud Encoding in Automotive Radar Perception
Organizations: Infineon Technologies AG, Austria · Graz University of Technology, Austria · Virtual Vehicle Research GmbH, Austria
Abstract
Radar represents a key modality for weather-robust perception. Unlike optical sensors such as camera and LiDAR, radar can penetrate fog, rain and snow due to its longer wavelength, and recent hardware advances have steadily improved its spatial resolution, thus making it increasingly relevant for autonomous systems. While processing raw radar data requires substantial memory computation, most existing work has focused on preprocessed point-cloud data similar to LiDAR. However, the sparse radar points and sensor-relative anisotropy make LiDAR-oriented encoders suboptimal. To address this problem, we present DRaGAN, a dynamic graph attention network for multi-radar point clouds. DRaGAN introduces a sensor-relative Fourier encoding of angle and range to capture fixed-aperture radar characteristics, gated short- and long-range hub nodes along each sensor's line of sight for efficient local-global message passing, and attention-guided dynamic rewiring that prunes weak edges and adds a bounded number of latent connections restricted to two-hop neighborhoods. This reduces the number of processed edges in deeper layers while preserving representation quality. On TruckScenes, DRaGAN improves test NDS by 10.10 points over the evaluated RadarGNN baseline and reduces the reported model FLOP count by 63.7%. This trade-off includes an increase in model latency from 40 to 53.16 ms and in trainable parameters from 1.2M to 1.7M on the evaluated implementations. Code and pretrained models are available at https://github.com/chr-is-tof/DRaGAN.
Figures & tables
| Method | Modality | NDS | mAP | mATE | mASE | mAOE | mAVE | mAAE | Split |
|---|---|---|---|---|---|---|---|---|---|
| CenterPoint [ 19 ] | L | 40.96 | 26.70 | 0.4089 | 0.3525 | 0.2769 | 2.7301 | 0.2008 | Test |
| PETR [ 27 ] | C | 12.14 | 2.22 | 1.1252 | 0.6863 | 0.6466 | 1.4987 | 0.5643 | |
| SpaRC [ 20 ] | R+C | 42.04 | 33.22 | 0.7174 | 0.3932 | 0.3545 | 0.7841 | 0.2077 | |
| HyDRa [ 21 ] | R+C | 22.37 | 12.78 | 0.7253 | 0.5441 | 0.7442 | 1.1804 | 0.3883 | |
| RadarGNN [ 8 ] | R | 10.78 | 6.98 | 0.8917 | 0.8087 | 1.1323 | 8.0028 | 0.5706 | |
| DRaGAN (ours) | R | 20.88 | 8.76 | 0.7277 | 0.6388 | 0.6442 | 1.0438 | 0.3391 |
| Metric | RadarGNN [ 8 ] | DRaGAN (ours) |
|---|---|---|
| Inference Time | ms | ms |
| FLOPs | ||
| MACs | ||
| Parameters |
| Metric | (a) | (b) | (c) | (d) | (e) | DRaGAN |
|---|---|---|---|---|---|---|
| NDS | 19.49 | 20.07 | 15.23 | 10.59 | 17.63 | 23.25 |
| mAP | 6.23 | 5.09 | 5.9 | 2.15 | 6.05 | 6.60 |
| mTPS | 163.75 | 175.25 | 122.80 | 95.15 | 146.05 | 199.50 |
Autonomous Driving Perception
Autonomous driving perception enables vehicles to detect and interpret road users, drivable space, and hazards using onboard and shared sensors.
See all 410 papers