*3.3. Simulation*

To validate the proposed method, simulations were performed with different scenarios and UAV numbers. Through the simulation, it was possible to verify the proposed approach, i.e., to plan the UAV tasks to map a catastrophic environment. Simulations were created with the following setup: CPU, Intel Xeon with 3.33 GHz 6 Core, 6 GB 1333 MHz DDR3 memory, and GPU ATI Radeon HD 5770 1024 MB.

There are several robot simulation environments, such as Open HRP [30], Gazebo [31], Webots [32] and Virtual Robot Experimentation Platform (V-REP) [33]. In this work, we chose V-REP, which has application programming interfaces (API) that allow communication with many programming languages. The proposed approach was implemented in MATLAB [34].

Figure 8 shows the simulation scenarios. Both were 10 × 10 m locations. Scenario 1 (Figure 8a) presents a place characterized by rooms with furniture that were knocked down, like an earthquake scene, while Scenario 2 (Figure 8b) is a place with passages; red dots represent fire spots.

**Figure 8.** V-REP simulation scenarios.

To perform exploration, the simulations made use of two and three similar UAVs. Figure 9 shows a used UAV. The UAV was equipped with an RGB-D camera, a thermal sensor, and a laser sensor. The laser sensor took a 0.5 cm radio to the honeycomb, so distance from a hexagon center to another was 1 m. For each scenario and each configuration (two or three UAVs), simulations were performed with the FIFO and Euclidean Distance algorithms.

**Figure 9.** UAV in simulation.

#### **4. Results**

Figure 10 shows scenarios merged with the honeycomb-map build. After the simulations were performed, it was possible to verify the displacements of each UAV within the generated map, as well as the order of honeycomb exploration by each UAV.

(**a**) Scenario 1 with honeycomb map. (**b**) Scenario 2 with honeycomb map.

**Figure 10.** Scenarios merged with honeycomb map.
