Enhanced 3D Object Detection via Diverse Feature Representations of 4D Radar Tensor

Recent advances in automotive 4-D radar have enabled access to raw 4-D radar tensor (4DRT), offering richer spatial and Doppler information than conventional point clouds. While most existing methods rely on heavily preprocessed, 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 3-D object detection framework that explicitly addresses the representation-level variability inherent in 4-D radar while preserving efficiency. Rather than assuming a single optimal radar representation, our method introduces a radar-centric multiteacher knowledge distillation (KD) framework, where multiple teacher models are trained on point clouds derived from diverse 4DRT preprocessing 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 AP3D and 9.5% in APBEV 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.