Rate-Adaptive LiDAR Point Cloud Streaming Over LEO Satellite Networks

Emerging Low Earth Orbit (LEO) satellite networks provide reliable connectivity in rural areas, enabling an autonomous agriculture comprising fleets of fully connected robots. Many agricultural robots are equipped with Light Detection and Ranging (LiDAR) sensors, which generate large, three-dimensional point clouds that are often offloaded to dedicated cloud-based computing centers for complex processing tasks. However, the produced data rates can reach hundreds of Mbit/s, overwhelming the uplink of current LEO satellite networks of approximately 20 Mbit/s, making a real-time transmission challenging. In this paper, we propose a novel solution that enables the real-time and rate-adaptive streaming of point clouds over bandwidth-restricted LEO satellite networks by introducing the concept of mini point clouds: a large point cloud is divided into smaller subsets that are individually compressed and transmitted. We develop a novel, real-time capable rate-adaptation algorithm leveraging Google Draco. It predicts optimal encoding parameters for each mini point cloud to maximize the point cloud quality under varying network conditions. It is configurable to prioritize computational speed or adaptation accuracy, and prevents congestion-induced packet loss by slightly under-utilizing the link. We extensively evaluate our algorithm on network traces from recent real-world Starlink measurements. It achieves real-time performance and near-optimal adaptation accuracy. Our solution is independent of the specific application and underlying network technology, making it broadly applicable to domains other than agriculture.