**Figure 10.**

*Industrial Robotics - New Paradigms*

developers in the world.

**3.3 Test in real scenario**

The first, Gazebo, is a 3D simulator for rigid bodies and robots, which offers the possibility to simulate precisely and efficiently robots in complex indoor and outdoor environments, with the ability to faithfully reproduce the real situation. The advantage of this tool is the presence of an easy programmable interface, but even more the fact of being an open source software with a strong active community of

Rviz is a suitable tool to view the 3D status of the robot and the performance of

The main purpose was to evaluate the functioning ability of the navigation algorithm. To do this, various simulations were carried out with different parameters, to

Once the algorithm and its procedure were validated in all virtual scenarios, the

The first test carried out using the drone in real scenario was operated in a facil-

the algorithms, to debug faulty behaviors and to record sensor data.

ity with technical characteristics described in **Table 2** and **Figure 10**.

3.Check communication link between UAV and ground station.

The types of tests that have been performed are divided into two categories:

**Table 3** shows the results obtained for type B condition and the relative absolute error calculated as the difference in Euclidean distance traveled by the UAV between the point of take-off and point of landing. The distances over which the tests were performed are respectively 10, 20 and 30 m, iterated 10 times in order to compare the error related to the odometry data. **Table 4** displays the average minimum and

**Stretch Height Width Length** a 2.15 2.40 11 b 2.10 2.35 30 c 1.90 2.40 5

• Type B: tests with on board LED lighting, in a dark environment

test the obstacle avoidance algorithm in every aspect.

behaviour of the system was tested in a real environment.

During the test a precise routine has been followed:

2.System power-on and lipo-battery connection on UAV.

• Type A: tests conducted in a lighting environment

1.UAV positioning at the beginning of the tunnel.

4.Execution of ROS launch file.

maximum error for each different test.

5.Set up mission parameters.

6.Start mission.

**158**

**Table 2.**

*Characteristics of the tunnel.*

*View of the tunnel.*


### **Table 3.**

*Absolute error in metres for each different test lengths.*


### **Table 4.**

*Minimum, maximum and average error.*

The minimum error obtained for the various ranges of distance tested is consistent with the results obtained in other recent works of the literature [13, 14]. At the same time, if we analyse the average error obtained by performing multiple consecutive tests for each range of distance, it can be seen that an improvement in the visual-inertial system is possible, although the system already guarantees great robustness in operation. An improvement could be obtained by using a different hardware, more performing IMU, and at the same time deepening the aspect related to the calibration of the camera in order to further succeed in decreasing the odometric error during navigation.

After conducting a series of test in facility, we definitively validate the result of the project and the system design in a different real tunnel (**Figure 11**). In this situation it was confirmed that the precision and reliability of the algorithms were enough to allow the system to navigate in total autonomy for at least a stretch of 100 metres.

**Figure 11.** *Tunnel inspection performed by the UAV during the test of system validation.*
