Dong-Hee Paek

Recent advances in automotive four-dimensional (4D) Radar have enabled access to raw 4D Radar Tensor (4DRT), offering richer spatial and Doppler information than conventional point clouds. While most existing methods rely on heavily pre-processed, sparse Radar data, recent attempts to directly leverage raw 4DRT incur high computational costs and limited scalability. To address these limitations, we propose a novel three-dimensional (3D) object detection framework that explicitly addresses the representation-level variability inherent in 4D Radar while preserving efficiency. Rather than assuming a single optimal Radar representation, our method introduces a Radar-centric multi-teacher knowledge distillation (KD) framework, where multiple teacher models are trained on point clouds derived from diverse 4DRT pre-processing techniques, each capturing complementary signal characteristics. These teacher representations are fused via a dedicated aggregation module and distilled into a lightweight student model that operates solely on sparse Radar inputs. Experimental results on the K-Radar dataset demonstrate that our framework achieves improvements of 7.3% in AP_3D and 9.5% in AP_BEV over the baseline RTNH model when using extremely sparse inputs. Furthermore, it attains comparable performance to denser-input baselines while significantly reducing the input data size by about 90 times, confirming the scalability and efficiency of our approach.

4D radar has attracted attention in autonomous driving due to its ability to enable robust 3D object detection even under adverse weather conditions. To practically deploy such technologies, it is essential to achieve real-time processing within low-power embedded environments. Addressing this, we present the first on-chip implementation of a 4D radar-based 3D object detection model on the Hailo-8L AI accelerator. Although conventional 3D convolutional neural network (CNN) architectures require 5D inputs, the Hailo-8L only supports 4D tensors, posing a significant challenge. To overcome this limitation, we introduce a tensor transformation method that reshapes 5D inputs into 4D formats during the compilation process, enabling direct deployment without altering the model structure. The proposed system achieves 46.47% AP_3D and 52.75% AP_BEV, maintaining comparable accuracy to GPU-based models while achieving an inference speed of 13.76 Hz. These results demonstrate the applicability of 4D radar-based perception technologies to autonomous driving systems.

Lane detection algorithm (LDA) is a crucial and necessary component for autonomous vehicles to ensure safe driving in various environments. Deep learning-based lane detection algorithms (DL-LDAs) have gained significant attention recently, and there have been a number of DL-LDAs, introduced in the literature, showing a continuous performance improvement in lane detection. In general, DL-LDAs are composed of pre-processing, lane feature extraction, lane detection head, and an optional lane fitting. For a systematic overview of various DL-LDAs, we provide detailed explanations for each functional component of DL-LDAs. Moreover, this paper presents the first survey to comprehensively analyze various DL-LDAs using camera and LiDAR, such as 2D (2-Dimensional) and 3D DL-LDAs using camera images and DL-LDAs using LiDAR point cloud or sensor fusion. In addition to the analysis, we present recent public lane detection benchmarks for DL-LDAs and discussions concerning technical issues that need to be addressed in future DL-LDA studies.

Accurate 3D multi-object tracking (MOT) is vital for autonomous vehicles, yet LiDAR and camera-based methods degrade in adverse weather. Meanwhile, Radar-based solutions remain robust but often suffer from limited vertical resolution and simplistic motion models. Existing Kalman filter-based approaches also rely on fixed noise covariance, hampering adaptability when objects make sudden maneuvers. We propose Bayes-4DRTrack, a 4D Radar-based MOT framework that adopts a transformer-based motion prediction network to capture nonlinear motion dynamics and employs Bayesian approximation in both detection and prediction steps. Moreover, our two-stage data association leverages Doppler measurements to better distinguish closely spaced targets. Evaluated on the K-Radar dataset (including adverse weather scenarios), Bayes-4DRTrack demonstrates a 5.7% gain in Average Multi-Object Tracking Accuracy (AMOTA) over methods with traditional motion models and fixed noise covariance. These results showcase enhanced robustness and accuracy in demanding, real-world conditions.

Drivable Area (DA) detection is crucial for autonomous driving. Camera-based methods rely heavily on illumination conditions and often fail to capture accurate 3D information, while LiDAR-based methods offer accurate 3D data and are less susceptible to illumination conditions. However, existing LiDAR-based methods focus on point-wise detection, so are prone to occlusion and limited by point cloud sparsity, which leads to decreased performance in motion planning and localization. We propose Argoverse-grid, a grid-wise DA detection dataset derived from Argoverse 1, comprising over 20K frames with fine-grained BEV DA labels across various scenarios. We also introduce Grid-DATrNet, a first grid-wise DA detection model utilizing global attention through transformers. Our experiments demonstrate the superiority of Grid-DATrNet over various methods, including both LiDAR and camera-based approaches, in detecting grid-wise DA on the proposed Argoverse-grid dataset. Grid-DATrNet achieves state-of-the-art results with an accuracy of 93.28% and an F1-score of 0.8328. We show that Grid-DATrNet can detect grids even in occluded and unmeasured areas by leveraging contextual and semantic information through global attention, unlike CNN-based DA detection methods. The preprocessing code for Argoverse-grid, experiment code, Grid-DATrNet implementation, and result visualization code are available at AVE Laboratory official git hub.

Drivable area (DA) detection is crucial for autonomous driving. Camera-based methods heavily rely on lighting conditions and often fail to capture accurate 3D information, while LiDAR-based methods offer accurate 3D data and are less susceptible to lighting conditions. However, existing LiDAR-based methods focus on point-wise detection, prone to occlusion and limited by point cloud sparsity, leading to decreased performance in motion planning and localization. We propose Argoverse-grid, a grid-wise DA detection dataset derived from Argoverse 1, comprising over 13K frames with fine-grained BEV DA labels across various scenarios. We also introduce Grid-DATrNet, a first grid-wise DA detection model utilizing global attention through transformers. Our experiments demonstrate the superiority of Grid-DATrNet over various methods, including both LiDAR and camera-based approaches, in detecting grid-wise DA in the proposed Argoverse-grid dataset. We show that Grid-DATrNet can detect grids even in occluded and unmeasured areas by leveraging contextual and semantic information through global attention, unlike CNN-based DA detection methods. The preprocessing code for Argoverse-grid, experiment code, Grid-DATrNet implementation, and result visualization code will be made available at https://github.com/kaist-avelab/grid-wise-DA.

Four-dimensional (4D) Radar is a useful sensor for the detection of surrounding three-dimensional (3D) objects under various weather conditions. However, since Radar measurements could be corrupted by noise, interference, multipath, and clutter, it is necessary to employ a preprocessing algorithm to filter out invalid measurements before the object detection with neural networks. In this paper, we propose RTNH+ that is an enhanced version of RTNH, a 4D Radar object detection network, based on two novel proposed algorithms. The first algorithm is the two-level preprocessing (TLP) algorithm that produces two measurement outputs from the same 4D Radar measurements at different filtering levels. The two outputs enrich the representation of the 4D Radar measurements and enhance object detection network performance. The second is the vertical encoding (VE) algorithm that effectively encodes vertical features of the surrounding objects in the TLP outputs. We provide details of the RTNH+ and demonstrate that RTNH+ achieves significant performance improvement of 10.14% in ${{AP}_{3D}^{IoU=0.3}}$ and 16.12% in ${{AP}_{3D}^{IoU=0.5}}$ over RTNH.The codes are available at https://github.com/kaist-avelab/K-Radar