Search-and-rescue missions often take place in environments where GPS signals are unavailable, unreliable, or severely degraded, such as collapsed buildings, underground facilities, tunnels, and dense urban areas. Ultra-Wideband (UWB) technology offers an effective solution by enabling accurate short-range distance measurements and relative localization between drones, fixed anchors, and rescue personnel. The proposed application combines autonomous drones, ESP32-S3 embedded controllers, and DWM3000 UWB transceivers to establish a cooperative localization network. Each drone can operate as a mobile anchor or tag and exchange ranging information with nearby devices. Single-Sided Two-Way Ranging (SS-TWR) can provide rapid distance updates with relatively low communication overhead, while Double-Sided Two-Way Ranging (DS-TWR) improves accuracy by reducing clock-offset errors.
During a rescue mission, the UWB measurements can be fused with visual–inertial odometry, LiDAR, or onboard computer vision to estimate each drone’s position in GPS-denied environments. The resulting coordinates can be displayed on a real-time map showing drone trajectories, anchor locations, detected victims, hazardous areas, and access routes. Multiple drones may also share their measurements through a cooperative communication network, increasing coverage and improving localization robustness.
UWB-enabled architecture can support:
Indoor drone navigation and trajectory tracking
Cooperative localization among multiple drones
Victim and rescuer position estimation
Relative distance measurement in low-visibility environments
Real-time rescue-scene mapping
Navigation assistance when GNSS is unavailable
SS-TWR and DS-TWR performance comparison
Integration with visual, inertial, LiDAR, and RF sensing systems
By combining UWB ranging with autonomous aerial platforms, the system can provide rescue teams with timely and spatially referenced information while reducing their exposure to unstable or dangerous structures. The application therefore represents a practical foundation for intelligent, collaborative, and resilient UAV operations in urban search-and-rescue scenarios.
Drone-to-Drone UWB Localization for Search-and-Rescue Operations
Drone-to-drone Ultra-Wideband (UWB) communication enables a group of unmanned aerial vehicles to estimate their relative positions in search-and-rescue environments where GNSS signals are unavailable or unreliable. Each drone is equipped with a UWB transceiver, such as the DWM3000, and an embedded controller for exchanging timestamped ranging messages with nearby UAVs. This cooperative architecture creates a mobile localization network that can operate inside collapsed buildings, tunnels, dense urban areas, and other GPS-denied environments. Relative distances between drones can be estimated using Single-Sided Two-Way Ranging (SS-TWR) or Double-Sided Two-Way Ranging (DS-TWR). SS-TWR requires fewer message exchanges and is suitable for rapid network updates, whereas DS-TWR compensates more effectively for clock-offset errors and generally provides greater ranging accuracy. For DS-TWR, the estimated time of flight can be expressed as
In a cooperative rescue mission, one UAV can operate as a temporary reference node while the other drones use inter-drone ranges to maintain formation and estimate their relative coordinates. When at least three geometrically distributed reference nodes or range constraints are available, multilateration can be applied to determine a drone’s position. The measurements may also be fused with visual–inertial odometry, LiDAR, optical flow, or inertial sensor data using an extended or unscented Kalman filter. This fusion reduces position drift and maintains navigation continuity when rubble, walls, smoke, or non-line-of-sight propagation degrades individual UWB links.
The resulting drone-to-drone network allows the swarm to distribute victim coordinates, obstacle information, local maps, and mission status among its members. A drone detecting a trapped person can associate the observation with its estimated position and relay the location to the rescue command station via neighboring UAVs. Other drones can then adjust their trajectories, improve scene coverage, or act as communication relays. UWB performance depends on antenna placement, synchronization, network scheduling, and propagation conditions. Metallic debris and reinforced concrete can produce multipath and non-line-of-sight bias. Robust operation therefore requires link-quality monitoring, outlier rejection, time-slotted ranging, collision avoidance, and confidence-weighted sensor fusion. With these measures, drone-to-drone UWB localization provides a resilient foundation for coordinated aerial mapping, victim localization, formation control, and navigation in GPS-denied search-and-rescue operations.
Integration of Reinforcement-Learning Agents with a UWB-Enabled Drone Localization System
The UWB-enabled localization architecture can be integrated with a reinforcement-learning (RL) agent to support autonomous drone navigation in GPS-denied search-and-rescue environments. In this framework, each drone operates as an intelligent agent that observes its local environment, estimates its position using UWB ranging, and selects navigation actions according to a learned control policy. This integration allows the drone to adapt its trajectory to obstacles, localization uncertainty, communication conditions, and mission priorities.
At time step t, the agent receives a state vector combining UWB and onboard sensor measurements:
The agent uses this state to select a control action:
The RL policy is optimized using a reward function that represents the principal objectives and constraints of the rescue mission:
Before the state is supplied to the agent, SS-TWR or DS-TWR measurements can be fused with inertial, visual, LiDAR, or optical flow data using an extended Kalman filter, an unscented Kalman filter, or a factor graph estimator. This sensor-fusion stage produces a more stable pose estimate and prevents noisy or non-line-of-sight UWB measurements from directly destabilizing the learned controller. Link-quality indicators and localization covariance should also be retained because they allow the RL agent to reason explicitly about measurement reliability.
In a multi-drone implementation, each UAV executes an individual policy while exchanging relative ranges, estimated poses, detected-victim coordinates, local maps, and link-status information. A centralized-training and decentralized-execution strategy can be employed: the agents have access to global mission information during training but make decisions using only locally available observations during deployment. The drones can consequently learn cooperative behaviors such as maintaining favorable ranging geometry, acting as temporary mobile anchors, relaying information, avoiding duplicated coverage, and approaching detected victims from different directions.
For safety-critical operation, the RL controller should not directly override the flight controller’s protective mechanisms. Its commands should pass through a safety layer that checks minimum obstacle clearance, geofencing limits, velocity constraints, battery condition, localization confidence, and communication availability. If UWB uncertainty exceeds an established threshold, the system can reduce speed, hover, reposition to restore line of sight, switch to visual–inertial navigation, or initiate a controlled return procedure. The integration of RL with cooperative UWB localization therefore creates an adaptive navigation architecture that balances exploration, victim search, positioning accuracy, energy consumption, and network connectivity. This approach is particularly valuable in dynamic and partially observable rescue environments where fixed navigation rules may be insufficient to address debris, multipath propagation, moving agents, and changing mission priorities.
Conclusion
The integration of UWB localization, reinforcement learning, sensor fusion, and autonomous UAV control provides a promising framework for search-and-rescue operations in GPS-denied environments. UWB ranging enables accurate relative positioning between drones and anchors, while onboard cameras, LiDAR, and inertial sensors improve navigation continuity under non-line-of-sight, multipath, and uncertain operating conditions. Combining these measurements through sensor fusion produces a more reliable state estimate for autonomous decision-making.
The reinforcement-learning agent enables the drone to adapt its trajectory to mission objectives, obstacles, localization uncertainty, energy availability, and communication quality. In multi-drone missions, cooperative UWB ranging can also support formation control, distributed mapping, victim localization, and information exchange. A safety layer remains essential for validating the agent’s commands and enforcing collision avoidance, velocity limits, geofencing, battery constraints, and emergency procedures.
Overall, the proposed UWB-enabled intelligent drone system can improve situational awareness, reduce localization errors, expand mission coverage, and limit rescuers’ exposure to hazardous areas. Future work should focus on real-world testing in complex disaster environments, non-line-of-sight error mitigation, multi-agent coordination, robust communication, energy-efficient planning, and validation of the complete system under realistic search-and-rescue conditions.
