Efficient On-Chip Implementation of 4D Radar-Based 3D Object Detection on Hailo-8L

TL;DR

This paper presents the first on-chip implementation of 4D radar-based 3D object detection on the Hailo-8L, enabling real-time, low-power autonomous driving perception with comparable accuracy to GPU systems.

Contribution

It introduces a tensor transformation method to adapt 5D CNN inputs for 4D tensor support on Hailo-8L, facilitating efficient on-chip deployment.

Findings

Achieves 46.47% AP_3D and 52.75% AP_BEV accuracy

Runs at 13.76 Hz inference speed

Maintains accuracy comparable to GPU-based models

Abstract

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.