Fernando Garcia 7/23/2021

This project aims to develop the algorithms that allow the navigation of two Unmanned Aerial Vehicles (UAVs) to locate an air pollutant source in a determined search area of 500x500 m2. The entire project consists of a proposed Navigation Strategy and 3 others strategies that are used to compare their results. These results will be published on the article named Equidistributed search+Probabilitybased tracking strategy to locate an air pollutant source with two UAVs (IEEE Access. The base algorithm used with each strategy is showed in the follow figure:

The simulated environment has the following characteristics:

• A “matched” experiment means that 1 UAV is flying at the same height of the pollutant source
• A “no matched” experiment means that none of the UAVs is flying at the pollutant source height
• Each UAV flies at different altitude
• The search area has no obstacles for UAVs
• The UAVs autopilot works with the MAVLINK navigation protocol
• The wind measurements were obtained from an anemometer located at 5 meters height (off-line)
• The simulation time for each experiment is 10 minutes (due battery restrictions)
• There is only one pollutant source in the search area
• The minimum detection level is 0.01 [ppm] (given by Official Mexican Standard NOM-038-ECOL-1993, for SO2). If a sensor measure exceeds this value it is considered as high pollutant concentration measure (named as “detection”)

The next table shows a summary of the main characteristics of each strategy and the results obtained in the experiments.

Table 1: The planning and control components

* improvement of measures = fist detection value - highest detection value

On first instance a logic supposition is that the distance to the source and the pollutant levels have a negative correlation on all experiments. This supposition is true when the height of the pollutant source matches with the flying height of a UAV, since its correlation index is -0.8. On the other hand, when no UAV flies at the same height of the pollutant source the variables are not correlated. The correlation index on this experiments is -0.15.

In this experiment, the source of contamination is at the same height as UAV1 (the green one) and UAV2 is one meter higher (6 meters above the ground). The green feather silhouette is the image of the z = 5 plane and represents the data that the UAV1 can measure. The blue feather silhouette is in the z = 6 plane and represents the data that the UAV2 can measure. The red area is the probability map.

Once a high measurement of pollutant is detected, the UAV fly on semicircular trajectories around the plume. With every high pollutant measurement the circle radius is reduced.

The third comparison strategy (or strategy 4) consists of a Brownian-like movement, developed by S. Zhang et. al. in Simulation implementation of air pollution traceability algorithm based on unmanned aerial vehicle. In this work, both phases have similar behavior. The difference for the exploitation phase is a rule that prevents the UAV to select a new fly direction if its current pollutant measure is higher than the previous one. Some modifications were required to adapt the original algorithm to our simulated platform:

1. In the exploration phase the x domain of the search map is divided for each UAV
2. The step of the UAVs is divided in half for the exploitation phase
3. The range of motion for both UAVs is restricted to be one step of the location of the best measurement taken

In this work, we presented an intelligent strategy to locate an air pollutant source on an outdoor area with two UAVs. This strategy was compared against the other three in simulated real-time experiments, where a dispersion-advection plume model was used. Unlike previous similar works, our research uses more realistic constraints on the UAVs platform (time of flight, ground speed, sensor sensitivity, communication coverage), in addition to experimenting with a very large search area and initial take-offs from different places. The proposed strategy uses an equidistributed search based on Hammersley sequences during the exploration phase. This allows the UAVs to cover different points of the search area, avoiding the repetition of sampling points. Additionally, the k-means grouping algorithm, TSP solver, and cubic spline algorithms are implemented in this phase to optimize and smooth the navigation. In the exploitation phase, the information taken by sensors is used to compute the probability of finding the pollutant source and redirect the search to better locations. In this phase, semicircular trajectories with decreasing radius are implemented.

The best results of the proposed strategy were obtained in the exploitation stage, showing final locations closer to the source and higher pollutant concentrations.

Future work will focus on implementing other bioinspired algorithms to explore efficiently the area. Also, planning is made to overcome the following disadvantages of the current work: in first instance, is necessary to add a wind model from acquired wind data to strengthen the probability map. The second improvement could be the replacement of the cubic spline with another path tracing algorithm. That algorithm must generate curved paths able to diminish the number of speed reductions in UAVs and increase the chances of having smoother navigation.

A. F. García-Calle, L. E. Garza-Castañón and L. I. Minchala-Avila, “Equidistributed Search+Probability Based Tracking Strategy to Locate an Air Pollutant Source With Two UAVs,” in IEEE Access, vol. 9, pp. 118168-118180, 2021, doi: 10.1109/ACCESS.2021.3099425.

AF projects: Providing high-quality and very reliable solutions for your science/engineering problems.