GLIDE: A Coordinated Aerial-Ground Framework for Search and Rescue in Unknown Environments

TL;DR

Empirical results demonstrate that explicit role separation across UAVs, coupled with terrain scouting and guided planning, improves reach time and navigation safety in time-critical SAR missions.

Abstract

We present a cooperative aerial-ground search-and-rescue (SAR) framework that pairs two unmanned aerial vehicles (UAVs) with an unmanned ground vehicle (UGV) to achieve rapid victim localization and obstacle-aware navigation in unknown environments. We dub this framework Guided Long-horizon Integrated Drone Escort (GLIDE), highlighting the UGV's reliance on UAV guidance for long-horizon planning. In our framework, a goal-searching UAV executes real-time onboard victim detection and georeferencing to nominate goals for the ground platform, while a terrain-scouting UAV flies ahead of the UGV's planned route to provide mid-level traversability updates. The UGV fuses aerial cues with local sensing to perform time-efficient A* planning and continuous replanning as information arrives. Additionally, we present a hardware demonstration (using a GEM e6 golf cart as the UGV and two X500 UAVs) to evaluate end-to-end SAR mission performance and include simulation ablations to assess the planning stack in isolation from detection. Empirical results demonstrate that explicit role separation across UAVs, coupled with terrain scouting and guided planning, improves reach time and navigation safety in time-critical SAR missions.

Figures (5)

• Figure 1: (Top Left) Real-world experimental visualization of the proposed end-to-end aerial-ground SAR framework. The test environment includes a U-shaped corridor at the center, with the victim located at the top. The dashed pink line indicates the UGV's planned trajectory prior to receiving terrain information from the terrain-scouting UAV, while the blue dot marks the footprint of the terrain-scouting UAV. The black dashed line indicates the ego-vehicle's planned trajectory after re-planning occurs. (Top Right) Ego-vehicle perspective as it approaches the U-shaped obstacles. (Bottom Left) Occupancy grid generated by our GLIDE method, smoothed over time, where dark gray regions denote detected non-traversable areas. The ego vehicle uses this map to plan avoidance maneuvers. (Bottom Right) Object detection results from aerial imagery captured by the terrain-scouting UAV. The detection module outputs bounding boxes for identified objects along with their associated confidence scores.
• Figure 2: The framework illustrates cooperative aerial-ground SAR operations: goal-searching UAVs survey the disaster site, detect and geolocate victims, and transmit compact messages to the UGV, while terrain-scouting UAVs track the UGV’s path to identify obstacles. The UGV fuses these inputs with local sensing and employs an A* planner to compute safe, time-optimal routes to the victims.
• Figure 3: Both UAVs run lightweight YOLOv11 detection pipelines, finetuned on the VisDrone dataset zhu2021detection and optimized with TensorRT, onboard the Jetson Orin Nano. The results are paired with the state information (retrieved from the flight controller via PyMavLink) of the UAV in order to georeference the victim positions. On the ground side, detections are ingested by the aerial mapping node, which fuses them with sensor inputs (LiDAR, RTK) to update a mid-level map. A planner node then computes safe trajectories using this map, while the control node sends low-level commands via the CAN bus to the drive-by-wire system. Together, these modules enable a coordinated, yet modular approach to achieving the objectives of a SAR-style mission scenario.
• Figure 4: (Left) One of the HolyBro X500 platforms utilized in the experimental studies as either a goal-searching UAV or a terrain-scouting UAV. (Right) The GEM e6 golf cart outfitted as the ego-vehicle for the real-world hardware demonstrations.
• Figure 5: Aerial-ground navigation in simulation. Left: Trajectories under GT (yellow), Local (blue), and GLIDE (green). Right: Sequential snapshots of one GLIDE trial, where the UGV (green) and UAV (purple) scouting progressively reveal obstacles (black) from the unexplored (dark grey), enabling replanning around them to reach the goal.