R2D2: Rural Road Detection Dataset

Multimodal sensor datasets play a crucial role in training machine learning algorithms for autonomous vehicles. While several such datasets exist, almost all of them deal with roads in urban settings. In this paper, we present Rural Road Detection Dataset (R2D2), a comprehensive collection of labeled point clouds for object detection and semantic segmentation of rural roads that aims to address this challenge. Our dataset includes a diverse range of rural environments and road types, providing a challenging environment for machine learning algorithms to learn from. With over 10,000 labeled point clouds captured from various locations, R2D2 is a valuable resource for researchers and practitioners working towards a safer and more efficient transportation systems in rural areas. We believe that our dataset will help to accelerate the development of autonomous driving in these remote areas and ultimately bring us closer to a future where all roads, no matter how rural, can be safely and efficiently navigated.

I. Introduction

Road transport plays a vital role in today's society. It has been the primary means of transportation in most countries, much more so in the United States. While 81% of America's population lives in and around cities, around 68% of all its roads are rural roads [1]. The low population density in these areas combined with the large areas they span causes a situation where there are more roads than people - 9,494 lane miles per 100,000 residents, as opposed to 1,056 in urban areas. This means that a resident in a rural community would have to travel more miles in order to access basic facilities. Fig. 1:

An example of a frame from R2D2.