Research Analyzer
← Back ICRA 2026

SpikeClouds: Streaming Spike-Based Processing of LiDAR for Fast and Efficient Object Detection

Michael Neumeier, Nael Fasfous, Bing Li, Axel von Arnim

PDF

AI summary

Key figure (auto-extracted from paper)
SpikeClouds enables low-latency, energy-efficient 3D object detection by streaming MEMS LiDAR data through a spiking neural network backbone.
LiDAR Spiking Neural Networks Streaming Processing 3D Object Detection Neuromorphic Computing MEMS Sensors

Problem

Full-scan LiDAR processing causes high latency and large memory footprints, while existing streaming methods only support rotating mechanical scanners, not efficient scanline-based MEMS LiDARs.

Approach

The method streams LiDAR scanlines as sparse spike-based representations called SpikeClouds, processes them with a spiking neural network backbone, and feeds the accumulated features into a CNN head for object detection.

Key results

  • Close to state-of-the-art detection accuracy on KITTI and JRDB22 datasets
  • 10% reduction in end-to-end latency on standard GPUs
  • 95% reduction in average memory footprint
  • 25× lower energy consumption on neuromorphic hardware

Why it matters

Provides a scalable, low-power perception pipeline for real-time robotics and autonomous driving systems.

Abstract

LiDAR sensors are used to provide three- dimensional information about the environment in many robotics applications. The information, accumulated in 3D point clouds, is first acquired by the sensor and then processed further, which leads to high end-to-end latencies and large memory footprints. Streaming approaches tackle this problem by processing partial point cloud data during scanning of the environment. In contrast to existing work that is limited to power hungry, rotating mechanical scanners, in this paper, we present a streaming method for more efficient scanline-based LiDAR sensors. We process the sequence of scanlines in form of SpikeClouds with a Spiking Neural Network (SNN) backbone and perform 3D object detection from the accumulated information using a Convolutional Neural Network (CNN) head. Our method achieves close to state-of-the-art detection performance on datasets KITTI and JRDB22 while reducing the end-to-end latency by 10% and the average memory footprint by 95% on standard GPU hard- ware. Additionally, when ported onto neuromorphic hardware, our backbone requires 25× less energy compared to reference backbones. SpikeClouds achieves fast and efficient environmental perception for robotic applications by streaming LiDAR to enable spike-based processing.

Index terms

Range Sensing Object Detection Segmentation and Categorization Neurorobotics

Related papers