- The paper demonstrates an effective balance between lightweight design and structural durability using FEA and impact simulations validated for SAR missions.
- It integrates advanced onboard perception with YOLOv8n-NCNN, achieving a 34% higher mAP and real-time target detection on a Raspberry Pi 5.
- A hybrid Light-Weight Incremental Greedy planner enables real-time dynamic rerouting, reducing the gap to optimal tour lengths by approximately 85%.
Integrated Design and Validation of a Micro-UAV for Dynamic Search and Rescue Operations
Introduction
This paper presents a comprehensive approach to the design, structural validation, and autonomous navigation of a sub-2 kg micro-UAV tailored for search and rescue (SAR) missions. The work addresses persistent challenges in the field: the trade-off between lightweight construction and structural durability, the computational constraints of onboard perception, and the need for real-time, adaptive route planning in dynamic environments. The proposed system is notable for its use of commodity hardware, open-source software, and a modular, impact-resistant airframe, all validated through rigorous simulation and experimental testing.
Structural Design and Validation
The UAV employs an X-frame quadrotor configuration, leveraging carbon fiber composites for the hub and arms to maximize the strength-to-weight ratio. The double-layer sandwich hub and cross-frame layout enable compact component integration and high bending stiffness. The landing gear, fabricated from TPU with a gyroid infill, is optimized for energy absorption and resilience on uneven terrain. A vacuum-formed PETG dome provides aerodynamic streamlining and protection for onboard electronics.
Finite Element Analysis (FEA) and LS-Dyna simulations confirm the structural integrity of the airframe under operational loads (40 N thrust, 20 N weight) and impact scenarios (15 m drop tests). The maximum displacement at the motor arm ends is 1 mm, with a peak stress of 0.4 MPa and a safety factor of 3.5. Modal analysis demonstrates that all primary vibrational and torsional modes are well above the motor operating frequency (166.67 Hz), eliminating resonance risk. Computational Fluid Dynamics (CFD) studies further refine the design, reducing the drag coefficient from 0.7 to 0.59 and validating a thrust-to-weight ratio of 2.
Onboard Perception and System Architecture
The perception pipeline is implemented on a Raspberry Pi 5, interfaced with a Pixhawk 6C flight controller via a 921 kbit/s serial link using ROS 2 XRCE-DDS. The system bypasses MAVLink parsing for real-time ingestion of region-of-interest (ROI) waypoints, enabling low-latency communication between perception and planning modules. Telemetry is streamed to QGroundControl for live visualization.
Object detection is performed using YOLOv8n, converted to the NCNN runtime for efficient inference on ARM-based hardware. Benchmarking against SSD-MobileNetV3-FPN Lite on a custom aerial dataset shows that YOLOv8n-NCNN achieves a 34% higher mAP (0.734 vs. 0.607) with only a 16 ms increase in latency, maintaining real-time performance at ~20 FPS for 640x480 input. The system incorporates robust fail-safes, including geofence-based Return-to-Home, RF link loss escalation, and battery-based auto-landing.
Dynamic Path Planning
The route planning algorithm alternates between a coverage (SWEEP) mode and a service (SERVICE) mode. The SWEEP mode executes a precomputed lawn-mower grid to guarantee area coverage. Upon detection of new ROIs (e.g., victims or obstacles), the system switches to SERVICE mode, selecting and visiting the three nearest ROIs using a nearest-neighbour heuristic. The planner is incremental: new detections are merged immediately, and the route is updated in real time, preventing stale waypoints.
The proposed Light-Weight Incremental Greedy (LIG) algorithm achieves a balance between computational efficiency and path optimality. In simulation, LIG reroutes in under 5 ms per update, closing approximately 85% of the gap to the globally optimal tour length (as computed by Concorde MILP for TSP), while avoiding the prohibitive latency (~70 ms per update) of full MILP solvers. Pure greedy nearest-neighbour heuristics are faster (~2-3 ms) but result in significantly longer tours, especially as target density increases.
Experimental Results
The integrated UAV system demonstrates robust performance in simulated and real-world SAR scenarios. Structural testing confirms the airframe's ability to withstand operational and impact loads, with foam-encased rods in the landing gear significantly improving damping and impact resilience. Aerodynamic refinements yield improved flight efficiency, and modal analysis ensures vibrational safety.
Onboard perception achieves real-time victim detection with high accuracy and low latency, validating the choice of YOLOv8n-NCNN for embedded deployment. The hybrid planning algorithm enables rapid, adaptive rerouting in response to dynamic mission conditions, with empirical results showing substantial reductions in detour distance compared to purely greedy approaches.
Implications and Future Directions
This work demonstrates that it is feasible to construct a structurally robust, perception-driven, and dynamically adaptive micro-UAV using only commodity hardware and open-source software. The integration of lightweight, impact-resistant materials with efficient onboard vision and planning pipelines enables practical deployment in SAR missions without reliance on expensive hardware accelerators.
The results suggest several avenues for future research. Enhancing detection robustness under low-visibility conditions (e.g., smoke, dust, or low light) remains a critical challenge. Coordinated multi-UAV operation, with decentralized planning and inter-UAV communication, could further improve coverage and responsiveness in large-scale SAR scenarios. Additional refinements in airframe design may yield further gains in portability and maneuverability, enabling operation in even more constrained environments.
Conclusion
The paper presents a validated, low-cost micro-UAV platform that addresses key limitations in structural durability, onboard perception, and dynamic route planning for SAR applications. The combination of carbon-fiber and TPU construction, efficient YOLOv8n-NCNN perception, and a hybrid greedy-optimal planner results in a system that is both robust and responsive. The approach provides a practical blueprint for future autonomous UAVs in real-world, resource-constrained environments, with clear potential for further enhancements in autonomy, robustness, and scalability.