A Real-Time Framework for Intermediate Map Construction and Kinematically Feasible Off-Road Planning Without OSM

TL;DR

The paper tackles off-road global path planning in regions where OpenStreetMap and Digital Elevation Model data may be unavailable, emphasizing real-time performance, memory efficiency, and kinematic feasibility. It introduces an intermediate map constructed in pixel space from raw geographic features, followed by a distance-map–driven Dijkstra pass, and a two-stage refinement that enforces kinodynamic constraints via selective Hybrid A* on locally sliced maps, aided by curvature smoothing and Voronoi-field costs. In large-scale tests, the approach achieves real-time operation with average times around 1.5 s and modest memory use (~1.5 GB) under challenging conditions, while delivering feasible, smooth trajectories suitable for search-and-rescue and agricultural tasks. The framework provides a practical pathway for robust off-road navigation in data-sparse environments and lays groundwork for integrating local planners and dynamic obstacle handling in future work.

Abstract

Off-road environments present unique challenges for autonomous navigation due to their complex and unstructured nature. Traditional global path-planning methods, which typically aim to minimize path length and travel time, perform poorly on large-scale maps and fail to account for critical factors such as real-time performance, kinematic feasibility, and memory efficiency. This paper introduces a novel global path-planning method specifically designed for off-road environments, addressing these essential factors. The method begins by constructing an intermediate map within the pixel coordinate system, incorporating geographical features like off-road trails, waterways, restricted and passable areas, and trees. The planning problem is then divided into three sub-problems: graph-based path planning, kinematic feasibility checking, and path smoothing. This approach effectively meets real-time performance requirements while ensuring kinematic feasibility and efficient memory use. The method was tested in various off-road environments with large-scale maps up to several square kilometers in size, successfully identifying feasible paths in an average of 1.5 seconds and utilizing approximately 1.5GB of memory under extreme conditions. The proposed framework is versatile and applicable to a wide range of off-road autonomous navigation tasks, including search and rescue missions and agricultural operations.

Figures (18)

• Figure 1: The proposed framework for off-road navigation. The initial step involves gathering essential raw data that encompasses the environment's features, including trees, buildings, bodies of water, and off-road trails. Once the raw data is preprocessed, an intermediate map is constructed as a preprocessing task. To accommodate user preferences and safety considerations, users are empowered to define restrictive and passable areas. The proposed global path planner leverages the constructed intermediate map and user-defined areas to generate real-time paths that adhere to kinematic constraints. Afterwards, we can integrate any off-road local planner for vehicle maneuvering.
• Figure 2: To map an uncharted terrain, we employed a T-Drone equipped with a Livox Avia LiDAR sensor.
• Figure 3: Georeferencing information and geographical features that are accepted as the input to the proposed off-road navigation framework. Initially, geographical features are converted into the pixel coordinate system and perform path planning and project planned path back to the original coordinate system (EPSG:4326)
• Figure 4: The proposed hybrid path planning algorithm leverages existing off-road trails and map data to determine a route between two positions. In this particular scenario, the start and target positions are located outside the off-road trails. However, the initially planned path is not the shortest possible. Algorithm \ref{['alg:find_closer_points']} is designed to identify the shortest path. As shown in Fig. \ref{['fig:point_selections_clusters']}, it successfully provides the shortest path for this example.
• Figure 5: Visual representation of Algorithm \ref{['alg:find_closer_points']} for finding closer poses on the off-road trails to the goal pose