RAVEN: Radar Adaptive Vision Encoders for Efficient Chirp-wise Object Detection and Segmentation

2026-04-06

AI Summary

This work addresses the high computational cost and latency of conventional frame-level radar perception by proposing a lightweight, streaming deep learning architecture tailored for FMCW radar. The method processes raw ADC data at the chirp level, preserving MIMO structure through per-receiver state-space encoders and introducing a learnable cross-antenna mixing module to generate compact virtual array features. Additionally, an early-exit inference mechanism based on latent state stability enables low-latency responses. Evaluated across multiple automotive radar benchmarks, this paradigm achieves state-of-the-art performance in both object detection and BEV free-space segmentation, simultaneously delivering high accuracy, low computational overhead, and end-to-end minimal latency.

Abstract

This paper presents RAVEN, a computationally efficient deep learning architecture for FMCW radar perception. The method processes raw ADC data in a chirp-wise streaming manner, preserves MIMO structure through independent receiver state-space encoders, and uses a learnable cross-antenna mixing module to recover compact virtual-array features. It also introduces an early-exit mechanism so the model can make decisions using only a subset of chirps when the latent state has stabilized. Across automotive radar benchmarks, the approach reports strong object detection and BEV free-space segmentation performance while substantially reducing computation and end-to-end latency compared with conventional frame-based radar pipelines.

Problem

• FMCW radar
• object detection
• free-space segmentation
• computational efficiency
• end-to-end latency

Innovation

• chirp-wise processing
• state-space encoder
• virtual-array feature
• early-exit mechanism
• FMCW radar perception