*4.2. Path Planning with Different Risk Combinations*

According to the result in the previous subsection, it can be seen that the model obviously mitigates the risk cost in path planning. However, the comprehensive risk model proposed considers three types of risks: pedestrian, vehicle, and building. The path planning results also are affected to an extent by the difference in risk models.

Therefore, further quantitative analysis is required to study the effects of different risk combinations on drone path planning and risk costs in urban environments. This section simulates and studies path planning in the above flight area with four risk combinations: (a) Group A considers three risks, (b) Group B considers pedestrians and buildings, (c) Group C considers buildings and vehicles, (d) Group D considers pedestrians and vehicles, and (e) Group E does not consider risks.

Figure 9 presents the effect of different risk combinations on path planning. Path A has a total risk cost of 6.20. Path E is the worst because it does not mitigate any risks, with 433.23% higher total risk cost than Path A. Path B and Path C have similar results, with Path C being 7.99% higher than Path B due to dense pedestrian areas being more relevant to buildings. The risk cost of Path D increases by 53.99% relative to Path A. Due to the gravity model, the distribution of pedestrians and vehicles is associated with buildings, and disregarding building risks leads to a subsequent small increase in pedestrian and vehicle risks, but this increase is significantly lower than Path B and Path C, where the corresponding risks are not considered.

**Figure 9.** Impact of different risk combinations in the environment on path planning.

The average path length is affected by the combination of risks, and Path E has the shortest length without considering risks. Considering all three risk types, the model proposed in this paper only increases the path length by 12.00% over Path E.

For the increase in path length, on the one hand, the 12.00% increase in path length is minimal compared to the 433.23% increase in risk. On the other hand, we add the constraint of drone energy consumption to the path planning model. Although the length of Path A increases, it still completes all customer service requirements and reaches the target point within the energy consumption constraint, indicating that the increase in path length is negligible.

The results show the path planning under different risk combinations to further understand the differences in the various types of risk costs of the path planning results while considering the risk combinations.

This paper investigates each type of risk cost (pedestrian risk cost R1, vehicle risk cost R2, and building risk cost R3) in the above five risk combinations. The results are shown in Table 4. Path C was planned without considering the risks associated with pedestrians in the environment. The drone path enters dense pedestrian areas, resulting in a pedestrian risk cost R1 of 15.48, which is higher than the case of Path A and Path B, where pedestrian risk is considered. On the contrary, the risk combination considered in Path B includes pedestrian risk, thus avoiding the area with high pedestrian risk costs. However, vehicle risk is not considered, resulting in a higher vehicle risk cost of 14.47. The exclusion of building risk in Path D leads to an increase in building risk by 452.63%.

**Table 4.** Split comparison of path risk costs.


The gravity model leads to an overlap of the three risk types, which is similarly demonstrated in the variation of the three types of risk cost. Path B ignores vehicle risk, while pedestrian risk and building risk increase respectively by 6.27% and 94.74%; Path C ignores pedestrian risk, but vehicle risk and building risk increase respectively by 16.92% and 573.68%; Path D ignores building risk, and pedestrian and vehicle risk increase respectively by 68.35% and 7.52%. Although Path D ignores the building risk, the drone path does not intrude into the high building risk zones due to the presence of pedestrian risk, so the building risk is reduced by 17.97% compared to Path B. The relevance of the

variation in different risk cost types also proves the importance of studying the integrated risk assessment model in this paper.

Path E presents that all three types of risk values are the highest among the five paths due to the correlation of various risk areas in the urban environment, such as the dense distribution of pedestrians and vehicles around the buildings. Therefore, the path planning results without considering any risk, the cost of all three risk categories is higher than the value of the corresponding risk category in any other combination.

For the mitigation effect of each type of risk, comparing Path E with Path A, it is shown that Path E in construction risk is 1.33 and Path A is 0.19, decreasing the risk by about 85.61%. Path E in vehicle risk is 15.11, and Path A is 2.66, decreasing the risk by approximately 82.40%. Path E for pedestrian risk is 16.62, and Path A is 3.35, decreasing the risk by about 79.85%. The total risk is reduced by approximately 81.25%. As a result, the model in this paper has a good mitigation effect on all three types of risks, and the proportion of the three risk reductions is kept at about 80.00%.

We can conclude that more risk sources in path planning can effectively mitigate the total path risk cost. This is because capturing more comprehensive risk sources is beneficial for avoiding more high-risk areas. It also further demonstrates the importance of our analysis and modelling for various types of elements threatened by drones in cities, which can guarantee the effectiveness of capturing risk costs in path planning.
