Collaborative firefighting robots bring aerial and ground-based automation together to detect hazards, map unfamiliar spaces and coordinate response tasks. In a proposed system combining an unmanned aerial vehicle (UAV) with an automated ground vehicle (AGV), each platform contributes different mobility and sensing capabilities. The aim is not to replace firefighters, but to provide early situational information and help deploy extinguishing material where the robots can reach.
The approach connects several demanding industrial automation problems: mobile-robot localization, perception in changing environments, wireless communication, task scheduling and safe control of physical equipment. A project demonstration showed how these elements can work together, while also revealing why integrated robotics must be evaluated as a complete system rather than as a collection of standalone components.
How collaborative firefighting robots divide the work
The UAV and AGV are designed to explore a fire scene together. The ground vehicle can approach and navigate at floor level, while the UAV can access elevated areas and investigate building openings. Both platforms use thermal cameras to identify hot regions that may indicate fire. When one robot detects a target, it shares the location so the other can factor that information into its own decisions.
A shared task database and scheduler provide the coordination layer. Tasks can be assigned according to factors such as a robot’s distance from a fire, battery condition and remaining extinguishing material. This is important because coordination is more than exchanging sensor readings: the system needs to determine which agent is best positioned and sufficiently resourced to act.
In the project’s example scenario, the AGV handles a fire on the ground floor while the UAV investigates a higher level. If one robot runs low on extinguishing material or cannot reach a target, the other may be assigned to assist when access permits. Each robot monitors its own task and reports progress, enabling the team to update its plan as conditions change.
Perception, mapping and navigation in an unknown space
Fire response depends on reliable estimates of where the robots are and what obstacles surround them. The AGV uses depth cameras to generate point clouds, which are converted into scan data for mapping and localization. The resulting occupancy and cost maps support path planning: the planner can identify traversable routes while accounting for obstacles.
Localization combines information from multiple sensors. In the AGV tests, tracking-camera data, an inertial measurement unit and wheel odometry were fused to improve pose estimates over longer runs. Wheel odometry alone can drift, for example when wheels skid or the ground is uneven. Sensor fusion can reduce that dependence on any one measurement source, although it still requires careful configuration and validation.
The UAV uses a thermal camera for fire perception and a depth camera for tasks such as detecting building openings. In the reported system, thermal-image processing used thresholding and morphological operations to isolate high-intensity regions. The robot then estimated the fire’s location from its pose and the detected region. This kind of method depends on the target’s thermal characteristics and visibility, so performance in a test setup should not be assumed to transfer unchanged to every real fire environment.
Navigation capabilities were not identical across the two platforms. The AGV had a mapping and planning pipeline, while the UAV’s mapping and obstacle-avoidance capabilities remained incomplete in the reported project. Its flight through an opening relied on a more constrained waypoint-based approach. That difference illustrates a key integration consideration: a coordinated system may share a mission objective without every agent having equivalent autonomy.
Closing the loop from detection to extinguishing
Detection is useful only if it leads to a controlled response. Once a robot is assigned a fire, it navigates toward the target, positions itself and activates its extinguishing hardware. The project used water tanks and pumps controlled through onboard computers and microcontrollers. Thermal-camera feedback informed the decision to engage the pump, linking perception software to a physical process.
That connection introduces practical engineering requirements beyond image recognition. The system must manage payload, power, communication and actuation, while avoiding unintended discharge. The project’s future-work plan identified water control as an area needing redesign: after deployment began, the mechanism could not reliably reserve and stop the remaining water. The team also noted that the robots could not determine how much water remained in their tanks.
These details matter in any automated response system. A useful task scheduler needs dependable status information, and a control system needs clear states for when to approach, spray, stop, request assistance or return. If resource sensing is unavailable, autonomy is constrained by uncertainty even when navigation and detection work as intended.
What the demonstration established—and what it did not
In a controlled demonstration using a mock building, the UAV and AGV detected represented fires, exchanged locations, navigated toward targets and deployed water. The team reported completion of a joint mission in 2 minutes and 41 seconds. The test also showed that the vehicles could cooperate on extinguishing tasks and avoid obstacles in the demonstration environment.
Those results are evidence of integrated functionality under the tested conditions, not proof of readiness for deployment in an uncontrolled structural fire. The report identified several unresolved limitations, including the UAV’s lack of obstacle avoidance and mapping, dependence on manual intervention for takeoff, landing issues, and the need to test opening detection in sunlight. Sensor and software failures also required attention: multiple USB-connected cameras could crash, and the system did not reliably detect every camera failure.
The development process offers a broader lesson for automation teams. Integration should begin early, because components that perform well independently may fail when combined. Simulation can support subsystem testing, but it cannot eliminate the need for physical trials. The project also highlighted the value of spare parts, fail-safe measures and schedule contingency when working with complex robotic hardware.
Practical implications for automation teams
UAV and AGV collaboration offers a compelling architecture for hazardous-environment response: aerial and ground agents can contribute different viewpoints, exchange target information and divide work based on location and available resources. Making that architecture dependable requires attention to shared coordinate frames, robust communication, fault detection, resource monitoring and safe actuation—not just fire-detection algorithms.
For plant managers, system integrators and engineering teams, the practical takeaway is to treat autonomous response as a staged capability. Validate sensing and navigation in representative conditions, test the full perception-to-actuation loop, and define what the robots do when information or equipment becomes unavailable. Collaborative firefighting robots can support faster assessment and targeted response, but their operational role must match the capabilities demonstrated and the limitations still being addressed.
Access Official Technical PDF