GND: Global Navigation Dataset with Multi-Modal Perception and Multi-Category Traversability in Outdoor Campus Environments

TL;DR

The paper addresses the challenge of outdoor global navigation by introducing GND, a large-scale, multi-modal dataset with multi-category traversability maps collected across ten university campuses. It provides a standardized data processing pipeline and demonstrates three useful applications—map-based global navigation, mapless traversal-aware navigation, and vision-based place recognition—across both wheeled and legged robots. The contributions include the first across-campus, long-range dataset with rich perception and traversability semantics, plus practical demonstrations of how such data enable global planning and recognition tasks. This dataset has significant impact for training and evaluating robust, semantics-aware global navigation methods in realistic outdoor environments.

Abstract

Navigating large-scale outdoor environments requires complex reasoning in terms of geometric structures, environmental semantics, and terrain characteristics, which are typically captured by onboard sensors such as LiDAR and cameras. While current mobile robots can navigate such environments using pre-defined, high-precision maps based on hand-crafted rules catered for the specific environment, they lack commonsense reasoning capabilities that most humans possess when navigating unknown outdoor spaces. To address this gap, we introduce the Global Navigation Dataset (GND), a large-scale dataset that integrates multi-modal sensory data, including 3D LiDAR point clouds and RGB and 360-degree images, as well as multi-category traversability maps (pedestrian walkways, vehicle roadways, stairs, off-road terrain, and obstacles) from ten university campuses. These environments encompass a variety of parks, urban settings, elevation changes, and campus layouts of different scales. The dataset covers approximately 2.7km2 and includes at least 350 buildings in total. We also present a set of novel applications of GND to showcase its utility to enable global robot navigation, such as map-based global navigation, mapless navigation, and global place recognition.

Figures (7)

• Figure 1: Traversability maps of the Global Navigation Dataset (GND): The legend shows five different categories of terrain traversability with different colors. Inset (a) shows the 3D satellite image of inset (b), an enlarged traversability map. GND contains multi-modal sensory data including 3D LiDAR point clouds (c), RGB images (d), and 360° images (e).
• Figure 2: Robot Setup: We use a Clearpath Jackal for dataset collection, which is equipped with various sensors, including 3D LiDAR, RGB camera, 360° camera, IMU, and GPS. It is capable of traversing diverse terrains, including pedestrian roads, roadways, off-road areas, ramps, and woods.
• Figure 3: Map-based Global Navigation using Multi-Category Traversability Map in GND: The white and yellow stars indicate the start and goal positions, respectively. The orange line shows the path of the wheeled robot, and the purple line shows the path of the legged robot. (a) and (b) shows Scenario 1, where there are stairs. (c) shows Scenario 2, where the robot is going to the right bottom of the figure. When the road is blocked and the legged robot can only go through the roadway due to narrow passages in off-road terrain. (d) shows Scenario 3, when the road is blocked and the robots have to go around the path using the off-road terrain.
• Figure 4: Partial Map of the University of Maryland: Inset (a) presents the processed point cloud map, featuring the robot's trajectory for data collection marked by a pink line; (b) illustrates the multi-category traversability map corresponding to the dataset shown in (a).
• Figure 5: T-MTG Generated Trajectories: T-MTG generates trajectories to cover traversable areas across three levels: basic, agile, and legged traversability levels. Each level imposes different constraints and utilizes different types of robots for navigation. Basic traversability level contains only pedestrian roads for all types of robots, agile traversability level includes both pedestrian roads and vehicle roadways for fast-moving wheeled robots, and legged traversability level has pedestrian walkways and off-road terrains for legged robots. For each cell, the left side displays the generated trajectories (red solid lines) overlaid on the robot's RGB view image, while the right side shows the generated trajectories (black dotted lines) overlaid on the cropped multi-category traversability map.