A Conflict Resolution Approach for Multiple Unmanned Aerial Vehicles

Abstract

The integration of Unmanned Aerial Vehicles (UAVs) into non-segregated airspace has raised critical conflict detection and resolution challenges, especially as UAV applications continue to expand. This paper presents a novel conflict resolution approach for multiple UAVs by segmenting the airspace into discrete altitude intervals. Two vertical maneuvering algorithms are proposed: one employing random selection and the other based on occupancy awareness. The simulation results demonstrate that the proposed method effectively resolves conflicts, reduces the likelihood of future encounters, and optimizes resource usage, contributing to safer and more efficient UAV operations in shared airspace.

1. Introduction

Unmanned Aerial Vehicles (UAVs) have seen widespread adoption across various sectors, including environmental monitoring, traffic surveillance, border patrol, and others. The commercial UAV market is also expanding rapidly, driven by advancements in technology and regulatory frameworks that facilitate access to non-segregated airspace [1].

This increased autonomy, however, presents significant challenges in ensuring safe separation between UAVs and other aircrafts. A conflict is defined as a situation where two or more UAVs experience a loss of minimum distance separation. In general, UAVs operate either in remote-controlled mode or autonomous mode, utilizing onboard sensors and GPS for navigation. While remote control offers centralized oversight, it may not be practical for large-scale operations, which necessitates robust conflict detection and resolution (CDR) mechanisms to detect and mitigate potential conflicts in autonomous mode.

Automating CDR enhances system safety and reliability by enabling UAVs to autonomously adapt to environmental changes, reducing reliance on central controllers. CDR comprises two core functionalities: conflict detection (CD) and conflict resolution (CR). In UAV operations, CD and CR are often designed separately before integration. CD involves using flight-state information from surrounding UAVs to detect and report potential future conflicts. Upon detection, a CR mechanism determines the appropriate maneuvers to resolve the conflict [1].

There are quite few UAV CDR approaches in the literature. Traditional CDR methods often focus on pairwise conflicts, which may not be sufficient as UAV traffic density increases. In addition, recent advancements in this field can be categorized into four primary approaches: geometric, optimization-based, artificial intelligence (AI)-driven, and hybrid methods.

Geometric methods remain foundational in UAV CDR, utilizing spatial relationships and trajectory analysis to detect and avoid conflicts. A notable advancement in this area is the probabilistic–geometric collision avoidance algorithm proposed by Lee et al. [2]. Optimization-based techniques model CDR as constrained optimization problems, aiming to find optimal conflict-free trajectories. A significant contribution is the work by Zhou et al. [1], which presents a novel approach for UAV CDR using tensor operations and an improved differential evolution algorithm. For AI and deep reinforcement learning CDR approaches, a notable development is the FRDDM-DQN algorithm [3], which combines Faster R-CNN for obstacle detection with a Data Deposit Mechanism to enhance training efficiency. In hybrid CDR methods, the strengths of geometric, optimization, and AI techniques are combined to address the limitations inherent in each.

In [4], a Straight-Line Conflict Detection and Alerting Algorithm, called SLIDE, is proposed as a distributed state-based alerting algorithm for 3D conflict detection. SLIDE is an efficient and practical CD approach that requires limited communication and exchanged information between UAVs. SLIDE has more realistic assumptions relevant to sensing and packet-loss and no prior knowledge is required to operate. It is based on analytical model and was validated extensively through simulation using OMNET++ [4,5]. However, SLIDE does not incorporate conflict resolution mechanism.

In this research, we extend SLIDE [4] to include conflict resolution in a multiple-UAV system. Our proposed approach is a novel CR approach that aims to address the existing limitations and adapt to unexpected flight conditions. The main contribution of this work is summarized as follows: (1) It efficiently and effectively positions UAVs after a conflict in real-time. (2) It proposes a novel mechanism for organizing the airspace into a set of altitude intervals to achieve clear separation between altitude levels. (3) It locally collects and calculates the UAV occupancy of each level. (4) It implements a rule-based maneuvering solution based on two factors: velocity and occupation of each altitude level. (5) Based on the above considerations, it implements two methods for conflict resolution: occupancy-based vertical maneuvering (OVM) and Random-based Vertical Maneuvering (RVM), which allows us to compare the performance accuracy and efficiency of the proposed model and resolution mechanisms.

The remainder of this paper is organized as follows: Section 2 presents the related work. In Section 3, we present the proposed CR approach, design considerations, and relevant assumptions. Section 4 discusses possible conflict scenarios. Section 5 presents a performance evaluation. Finally, Section 6 concludes with findings and future work.

2. Related Work

Research on conflict detection and resolution has gained an increasing interest in recent years. In this section, we review a number of existing and relevant approaches.

CDR approaches for UAVs are different from manned aircraft in many aspects, including communication model, flight regulation, traffic management, load balancing, maneuverability, and pilot mode. In addition, most UAVs fly at low altitude, which is more complex [6].

A recent survey on many unmanned aircraft traffic management systems is presented in [7]. It classifies the methods into two categories: (1) classical (non-learning) methods that typically rely on pre-defined algorithms or rules for UAVs to avoid collisions and (2) artificial intelligence (AI)-based methods using machine learning (ML) and more specifically reinforcement learning (RL). Such methods enable UAVs to adapt to their environment, autonomously resolve conflicts, and exhibit intelligent behavior based on their experiences [7].

In [8], Yang et al. proposed a cooperative, centralized solution in which CDR is formalized as a nonlinear optimization to minimize maneuver costs. It uses stochastic parallel gradient descent (SPGD) and an interior-point algorithm. In [9], Sislak et al. proposed a pairwise, cooperative, decentralized approach for CD and collision avoidance. A conflict resolution for a multiple-UAV system based on graph theory is also proposed in [6]. It presents a short-term CDR problem with cooperative heading control in fixed-wing UAVs. Razzaq et al. [10] presented a novel technique to find new routes for UAVs using graph theory.

Most of the existing CDR mechanisms suffer from a number of design and performance issues, including considering uncertainties; handling multiple conflicts, especially in real-time; lack of coordination; high computational and communication requirements; robustness to failure or performance degradation; consideration of environmental hazards; and certification requirements [11]. Therefore, many CDR approaches were proposed to achieve efficiency in terms of minimizing the overhead caused by UAV maneuver to resolve conflict.

In this work, we propose a conflict resolution approach in a multiple-autonomous-UAV system. It extends SLIDE [4,5], which is a distributed, state-based conflict detection approach in a multiple-UAV environment. SLIDE is adapted as a baseline for our approach because it views the intermediate future trajectory as a straight line and does not require a predetermined trajectory planning. It is also adaptive, as it allows UAVs to adapt their flight trajectories as needed. In addition, SLIDE is an accurate and efficient mechanism, as it requires limited communication and allows for the early detection of conflict.

Our proposed approach is a novel CR approach that aims to adapt to unexpected changes and frequent conflicts in a multiple-UAV environment. It presents a rule-based CR mechanism that considers velocity and occupation of each altitude level. To resolve conflict and avoid future conflicts, our approach includes a unique mechanism to virtually organize the airspace into a set of altitude intervals in order to achieve clear separation between altitude levels. It is also adaptive and context aware, as it locally collects and calculates the occupancy of each level in order to balance the number of UAVs between levels and avoid future conflicts. Next section presents in detail the design of the proposed approach.

3. The Proposed Conflict Resolution Approach

This paper proposes a novel CR approach for multiple UAVs, extending the Straight-Line Conflict Detection and Alerting Algorithm (SLIDE) [4] by integrating a CR mechanism. SLIDE has a number of interesting features: (1) it allows for a group of UAVs to share a common airspace, (2) it views UAVs’ trajectories in the near future as straight lines, and (3) it only relays the position and velocity information that are exchanged periodically between UAVs. Accordingly, (4) it does not require any previous knowledge of UAVs trajectory. In SLIDE, (5) UAVs can dynamically change their trajectories based on the flight situations and the surrounding environment. In addition, SLIDE (6) does not require any remote visual inspection (RVI) tools, such as high-resolution cameras. (7) SLIDE has an ability to model future trajectories without predefined plans, allowing UAVs to adapt to unexpected events. Those features are inherited and maintained in our CR approach, as SLIDE is considered the baseline CD mechanism to our approach.

Our proposed CR approach resolves conflicts by identifying separate altitude intervals and implementing vertical maneuver. Two algorithms are introduced: random vertical maneuver, which selects alternate altitudes randomly, and occupancy-based vertical maneuver, which considers the number of UAVs occupying nearby altitude levels to minimize congestion.

In the remaining parts of this section, we clarify CD using SLIDE and related assumptions. Then, we present our proposed CR approach, explain the mixed mobility model designed in our approach, and finally address the possible conflict scenarios.

3.1. Conflict Detection Phase

Before presenting the design of our approach, we present the assumption related to the conflict detection phase using SLIDE as an underlining framework and conflict detection mechanism for our approach.

SLIDE is a distributed, state-based alerting algorithm for three-dimensional CD among multiple UAVs. One of its major advantages is that it has low message and space complexity. In SLIDE, the communication between multiple UAVs is infrequent. Moreover, the exchanged STATE messages between UAVs only includes position and velocity vectors. Each UAV detects future conflicts using its own state information and neighboring UAVs’ state messages. Each UAV views the intermediate future trajectory in a straight line and therefore does not require a predetermined trajectory plan, which allows UAVs to adapt their flight trajectories as needed. The mobility model adopted by SLIDE is the common random waypoint model (RWM) [12], which is used to generate three-dimensional movement for multiple UAVs.

In the proposed algorithm of SLIDE, each UAV linearly projects its current flight-state information and reads STATE messages from neighboring UAVs to derive conflict parameters and decide imminent conflict risk. If the distance separating two UAVs is less than a predefined threshold, each UAV detects conflict and future loss of separation (unit) within a period ($t_{in}$) and calculates its end time ($t_{out}$). Accordingly, both UAVs register conflict and raise conflict alert, which is sent to a remote-control station or invoke a CR functionality. In our work, it invokes one of the implemented CR methods: RVM or OVM.

SLIDE includes tuning parameters such as broadcast cycle ($T_b$) and look-ahead time ($T_l$), which are used to determine whether a UAV is encountering a loss of separation within $T_l$. The broadcast cycle ($T_b$) also determines the time at which each UAV will broadcast its periodic STATE message. For instance, the STATE message of a specific UAV $\left({UAV}_a\right)$ comprises the current position vector ($P_a$), velocity vector ($V_a$), and protected-zone radius ($R_a$).

In conflict detection between two UAVs, A and B, once a UAV B ($\left({UAV}_b\right)$) receives a STATE message from a neighboring UAV A, the following actions are invoked:

• First, B estimates the time of closest approach ($t_{ca}$):
• If $t_{ca}<0$, then the two UAVs are traveling away from each other, and no future conflict is predicted.
• Otherwise, if $t_{ca}\ge 0$, then the two UAVs are traveling toward each other and expected to reach their closest point.

The distance of closest approach ($d_{ca}$) is used as follows:

• If $d_{ca}>(R_a+R_b)$, then no future conflict is predicted.
• If $d_{ca}\le (R_a+R_b)$, then a conflict is predicted within the period ($t_{in}$).
• If $V_{∆}=0$, then both UAVs are at the same distance of separation and $t_{in}$ will be zero.
• If $V_{∆}\ne 0$, then $t_{in}$ exists and must be calculated.
• If $0\le t_{in}\le T_l$, then conflict is registered, and alert is raised, conflict is registered, and our conflict resolution approach is invoked to resolve the conflict.

3.2. Conflict Resolution Phase

We extended SLIDE by introducing CR to resolve the detected conflicts and minimize future conflicts. Our proposed CR mechanism includes a rule-based maneuver solution that can help multiple UAVs adapt to different unexpected events. The proposed approach organizes the airspace into separate altitude intervals to have a clear separation between altitude intervals. It also allows each UAV to locally collect information about the occupancy of each altitude level.

As explained earlier, SLIDE uses the random waypoint model as a mobility model for multiple UAVs. We extend this model to generate both two- and three-dimensional movement. In designing our approach, we consider three major assumptions. First, we assume that the airspace is composed of different separate altitudes intervals. Second, we assume that each UAV knows its exact flight mode and altitude level. Third, we assume perfect sensing and that all packets will arrive successfully without any loss.

Initially, each UAV is launched at a random position in three-dimensional space. Once a UAV detects a conflict with another UAV, it makes a vertical maneuver to modify its current altitude in order to resolve conflict. The new altitude level will be fixed, and it will continue to fly in the two-dimensional (2D) plane. The choice of the new altitude level and which UAV to enter the 2D flying mode is explained next.

Figure 1 shows the flowchart of the proposed approach. The main steps are explained as follows: Upon detecting a conflict by SLIDE, our conflict resolution approach invokes one of our own designed maneuvering algorithms: (1) the random-based maneuver or (2) occupancy-based maneuver. The selection of RVM or OVM is configured manually based on the prior setting that determines CR mode. In both algorithms, we assume the airspace is organized into different altitude intervals called levels. Assuming that the available airspace is divided into m levels, when a conflict is detected, the movements of each involved ${UAV}_i$ will be restricted to a specific level L_i, creating 2D movement within that level L_i instead of the default 3D movement employed by SLIDE. Accordingly, if a ${UAV}_i$ is flying in 2D mode, its future maneuvers are restricted to that chosen level L_i. Determining the level L_i depends on the invoked maneuvering algorithm.

Moreover, in our approach, we extended the format of STATE messages, defined by SLIDE, to include a mobility mode indicator $I_i$ of ${UAV}_i$. This extension is considered to be lightweight, as it introduces a fixed number of numerical fields (flags). For instance, in our approach, the STATE message of a ${UAV}_a$ includes $A(P_a, V_a, R_a, I_a)$, where the variables are defined as follows:

-

$P_a$ is the position vector;

-

$V_a$ is the velocity vector;

-

$R_a$ is the radius of the protected zone;

-

$I_a$ is a mobility mode indicator.

The default value of the indicator $I_a$ is zero, which indicates that ${UAV}_a$ is flying in the default 3D mode. Accordingly, A can perform vertical maneuvers as needed. However, once ${UAV}_a$ experienced conflict, the value of its indicator $I_a$ is changed to 1 in order to indicate that a is flying in the 2D plane, and its future movements (maneuvers) are restricted within the determined level $L_a$.

The vertical maneuvers that UAVs can perform are either ascent ($m_{up}$) or descent ($m_{down}$). However, each UAV can maneuver once in each conflict. For instance, upon registering a conflict alert using SLIDE between ${UAV}_a$ and ${UAV}_b$, the conflict resolution algorithm is started at time t within ${\mathrm{t}}_{in}\le t<{\mathrm{t}}_{out}$, where ${\mathrm{t}}_{in}$ and ${\mathrm{t}}_{out}$ are calculated as previously determined by SLIDE.

3.2.1. Random-Based Vertical Maneuver (RVM) Algorithm

The RVM algorithm allows UAVs to randomly select an alternative level to resolve conflict. In the RVM algorithm, when a conflict is detected between ${UAV}_a$ and ${UAV}_b$ within a time t, ${UAV}_a$ is going to randomly choose to shift from its current level $L_a$ to a randomly chosen upper or lower level. Once ${UAV}_a$ chooses an alternative level ${L^{\prime }}_a$ $\mathrm{t}\mathrm{o} \mathrm{a}\mathrm{v}\mathrm{o}\mathrm{i}\mathrm{d} \mathrm{c}\mathrm{o}\mathrm{n}\mathrm{f}\mathrm{l}\mathrm{i}\mathrm{c}\mathrm{t},$ ${UAV}_a$ will not be able to change its new level. Accordingly, ${UAV}_a$ will continue to fly within altitude level ${L^{\prime }}_a$ until it reaches its destination. As a result, the mobility of ${UAV}_a$ will be projected from 3D space to a specific 2D plane. Algorithm 1 below shows the pseudocode of RVM algorithm.

Algorithm 1. Pseudocode of RVM algorithm

Input: ${UAV}_a$ movement indicator $I_a$, $t_{in}$, $t_{out}$, conflict alert Output: random alternative level ${L^{\prime }}_a$ If conflict alert = 1 then If $t_{in}\le t<t_{out}$ then If $I_a$ = 0 then randomly choose alternative level ${L^{\prime }}_a$ return ${L^{\prime }}_a$ End

3.2.2. Occupancy-Based Vertical Maneuvering (OVM) Algorithm

The OVM algorithm tries to perform vertical maneuvers to the least occupied altitude level. OVM aims to balance the distribution of UAVs among the available altitude levels. Based on our defined criteria, a UAV experiencing a conflict is going to shift to the nearest less-occupied level.

In order to allow UAVs to cooperatively calculate the occupancy of different levels, we propose a technique to propagate the occupancy information. The mobility nature of a multiple-UAV system requires frequent exchange of information between UAVs regarding topology and neighborhoods. In SLIDE, STATE messages are broadcasted within a limited cycle. In our approach, we propose exchanging an additional small message called AFFIX that announces affixing UAV A at a certain altitude level ${L^{\prime }}_a$ and changing its flying mode to 2D. The AFFIX message is a simple message that includes the ID of UAV A that originates it (${UAV}_a$) and its new level ${L^{\prime }}_a$. Upon receiving the AFFIX(${UAV}_a$, ${L^{\prime }}_a$.) message by neighboring UAVs, they will notice the increase in the number of affixed UAVs at that certain level. This AFFIX message will be flooded in the network for a certain number of hops controlled by a time-to-live counter.

In response to receiving the AFFIX message, each ${UAV}_i$ is going to maintain two local vectors: ${OC}_i$ to count the occupancy of each level and ${ON}_i$, which records the level ${L^{\prime }}_a$. to which a neighboring ${UAV}_a$ is fixed at after encountering a conflict. Initially, both vectors are initiated with zeros; then, the required information is collected progressively. The vector $O{C}_i$ is of size $m\times 1$, where m is the number of levels, while the vector ${ON}_i$ is of size $n\times 1$, where $n$ is the number of UAVs. Both vectors are initialized to zero and updated based on the received AFFIX messages.

Algorithm 2 below shows the pseudocode of the OVM algorithm. When a conflict is detected between ${UAV}_a$ and ${UAV}_b$ at time t, each UAV is going to decide about its vertical maneuver based on the following rules: First, the UAV with the larger ID value, the newer UAV, is going to perform the maneuver. To simplify, we limited the comparison to the ID value of the conflicted UAVs. However, it can be extended to consider the energy level or other factors. If ${UAV}_a$ is chosen to perform the maneuver, it searches the ${OC}_a$ vector to choose the least-occupied level. If there exist multiple levels with identical least-occupied value, OVM choses the nearest level.

Algorithm 2. Pseudocode of OVM algorithm

Input: STATE messages of ${UAV}_a$ and ${UAV}_b$, occupancy vectors O$C_a$, O$C_b$, $t_{in}$, $t_{out}$, conflict alert

Output: calling algorithm III for A or B, alternative level $i$, AFFIX message.

If ${UAV}_a$ is not in 2D

If ${ID}_a>{ID}_b$

Then ${UAV}_a$ can performs vertical maneuver

${L^{\prime }}_a=$ Find least-occupied and nearest level in ${OC}_a\left[i\right]$

broadcast AFFIX (${ID}_a$, ${L^{\prime }}_a$)

${UAV}_a$ performs maneuver within ${L^{\prime }}_a$

EndIf

ElseIf ${UAV}_b$ is not in 2D

Then ${UAV}_b$ can performs vertical maneuver

${L^{\prime }}_b=$ Find least-occupied and nearest level in ${OC}_b\left[i\right]$

broadcast AFFIX (${ID}_b$, ${L^{\prime }}_b$)

${UAV}_b$ performs maneuver within ${L^{\prime }}_b$

ElseIf both UAVs are in 2D, OVM reports that it cannot handle this conflict.

End

Once ${UAV}_a$ or ${UAV}_b$ choose an alternative level using RVM or OVM, they randomly generate movement within their new altitude level.

4. Conflict Scenarios

In this study, we mainly considered two conflict scenarios: crossing and heading scenarios. We distinguished between two types of mobility modes: 3D and 2D. However, to simplify, we focused on 3D–3D conflicts occurring between two UAVs both in 3D mode (both UAVs can change their altitude level to maneuver) and 3D–2D conflicts occurring between one UAV in 3D and another UAV in 2D mode (which means that one of the conflicted UAVs is in 2D mode and cannot change its altitude level to perform a maneuver). The 2D–2D conflicts will be discussed in future work.

4.1. Crossing vs. Heading Scenarios

Figure 2 shows the crossing scenario that occurs between two UAVs, ${UAV}_a$ and ${UAV}_b$, while they are heading to their destination. Figure 3 shows the heading scenario where the conflicted UAVs are at the same level. The presented conflicts in Figure 2 and Figure 3 can occur in 3D–3D and 3D–2D scenarios.

4.2. Conflicts Resolution in 3D–3D Scenarios

UAVs are initially launched in 3D flying mode. This scenario focuses on the conflict between two UAVs flying in 3D mode. The presented conflicts in Figure 2 and Figure 3 can occur in 3D, which represents the default flying mode. In this case, either RVM or OVM algorithms will be executed to resolve the conflict and decide the new altitude level for both UAVs in RVM or for the selected UAV based on the OVM algorithm.

4.3. Conflicts Resolution in 3D–2D Scenarios

The resolution introduced by the CR function could lead to other conflicts. The crossing conflict is the most common conflict. We considered the diverging conflict scenario, which is one type of crossing scenarios defined in [13]. A diverging conflict occurs when two UAVs that share the previous constraints come into conflict in different constraint. If a ${UAV}_b$ that is flying in 2D mode within a level ${L^{\prime }}_b$ has a conflict with another ${UAV}_a$ flying in 3D mode, then ${UAV}_a$ will perform the required vertical maneuver in order to resolve the conflict with ${UAV}_b$.

4.4. Resolving 2D–2D Conflicts

There are many possible scenarios for conflicts between two UAVs flying in 2D mode. Considering the heading conflicts between two UAVs at the same level, Figure 4 and Figure 5 show that a heading conflict can occur between UAVs of the same velocities or different velocities and when UAVs are moving toward each other or in the opposite direction.

It is important to highlight that resolving the heading conflicts in 2D is possible by manipulating velocities and/or adjusting the altitude value within the same level. However, resolving 2D–2D conflicts will be considered in future research work.

5. Performance Evaluation

This section discuss the performance evaluation, including the details of simulation setup, performance factors, and simulation results.

5.1. Simulation Tool and Setup

The implementation of our approach was performed using OMNET++ simulator version 6.0. In general, we employed the same simulation setup of SLIDE with considered variations. We considered N number of UAVs flying for 30 min in restrained three-dimensional space represented by the size 500 m $\times$ 400 m $\times$ 30 m. We assumed that UAVs move at constant speed toward predetermined destinations. We assumed the UAVs have an equal communication range. Communication between UAVs was conducted using IEEE 802.11 protocol [14]. We assumed the altitude levels were of the same size. The considered number of UAVs (N) included 20, 25, 30, 35, and 40 UAVs. The velocity v was assumed to be 3m/s and communication range = 25 m. Number of levels L in the airspace was assumed to be 20. We assumed that UAVs can also fly for more than 30 min. The presented results below represent an average of 30 repetitions to reach the required level of confidence. Accordingly, no. of conflicts in the following subsection presented the average for 30 repetitions. To simplify, we rounded the values of no. of conflicts.

5.2. Performance Measures

In our approach, we considered the performance measures defined by SLIDE, including the following:

• No. of actual conflicts;
• No. of predicted conflicts;
• No. of accurate alarms;
• No. of missed alarms;
• No. of false alarms;
• Average maneuver time.

We also introduced additional performance measures that are defined as follows:

• Three-dimensional–three-dimensional conflicts: conflicts detected between two UAVs that are both in three-dimensional space.
• Three-dimensional–two-dimensional conflicts: conflicts detected between two UAVs one in three dimensions and the other in two dimensions.
• Two-dimensional–two-dimensional conflicts: conflicts detected between 2 UAVs in two-dimensional plane.
• Total handled conflicts: the portion of conflicts resolved by our approach.
• Maneuvering energy consumption: the total amount of dissipated energy by all UAVs caused by vertical maneuvering as a result of encountering a conflict.

We implemented our energy model to calculate energy dissipation using the model presented by M. Hwang et al. [15]. Mainly, it computes energy dissipation during the maneuver as follows:

Energy Dissipation = (P_maneuver − P_hover) × t_maneuver

where

P_maneuver = Power during the maneuver (W);

P_hover = Hovering power (W);

t_maneuver = Duration of the maneuver (s).

We calculate energy dissipation due to altitude changes as follows:

Energy Dissipation = (P_hover,new − P_{hover,sea level}) × t_maneuver

where

P_hover,new = Hovering power at the new altitude (W);

P_{hover,sea level} = Hovering power at sea level (W).

5.3. Simulation Results

In this section we discuss the results of simulating our proposed approach. We compare the performance of RVM and OVM mechanisms. We extended the simulation framework developed by [4] to introduce CR. For each considered simulation scenario and duration, our simulator runs SLIDE to detect and count the no. of encountered conflicts, as shown in in

Table 1. Conflict detection using SLIDE.

	No. of UAVs (N)
	20	25	30	35	40
Velocity = 3 m/s	8.93	8.93	20.4	20.4	35.93
Velocity = 5 m/s	13.8	22.07	32.6	45.47	56.67

. For this experiment, broadcast cycle $T_b=2$ and the considered velocity values are 3 m/s and 5 m/s. As the table shows, increasing the no. of UAVs increases the no. of conflicts in the airspace. Moreover, incrementing the velocity increases the no. of conflicts. As explained earlier, SLIDE only provides conflict detection without any sort of resolution.

In contrast to SLIDE, RVM tries to resolve conflicts with minimal coordination and calculation cost. In RVM, both conflicting UAVs are taking random vertical maneuvers, which maximizes energy consumption as a result of maneuvering both UAVs. Moreover, the random choice of levels might resolve the current conflicts but increases future conflicts.

Table 2. RVM algorithm results for different number of UAVs with v = 3.

N	Detected Conflicts	Handled Conflicts	3D–3D Conflicts	3D–2D Conflicts	2D–2D Conflicts	Maneuvering Energy Consumption
20	11	8	3	6	3	187.96
25	15	11	4	8	3	248.57
30	23	14	5	11	7	307.62
35	33	19	7	14	11	467.75
40	46	24	9	19	18	594.97

shows the results of conflict detection and resolution when the RVM mechanism is employed for the same simulation scenario. While different conflicts are detected using SLIDE, our conflict resolution study focuses on resolving 3D–3D and 3D–2D conflicts using the vertical maneuvering. The table shows the total no. of 3D–3D and 3D–2D conflicts and the overall conflicts that are successfully handled by RVM. Moreover, 2D–2D and 3D–2D conflicts represent the conflicts introduced by our resolution mechanism after moving UAVs to 2D. The maneuvering energy consumption represents the energy consumed, in (nj), by all UAVs as a result of maneuvering both conflicted UAVs.

As velocity has a significant impact on increasing the number of conflicts and maximizing resolution complexity, we studied the performance of our approach with different velocity.

Table 3. RVM performance when v = 5 m/s.

N	Detected Conflicts	Handled Conflicts	3D–3D Conflicts	3D–2D Conflicts	2D–2D Conflicts	Maneuvering Energy Consumption
20	15	10	4	7	4	230.85
25	28	16	6	11	10	368.92
30	39	19	7	15	17	429.53
35	58	26	9	22	28	595.42
40	76	31	10	27	39	780.63

shows the detailed performance results of RVM when velocity v = 5 m/s. Increasing the velocity increases the number of conflicts.

Figure 6 compares vertical maneuvering energy consumption in RVM when velocity varies from 3 to 5. Similar to SLIDE, the number of conflicts is proportional to velocities. The figure shows that the increase in energy consumption is almost linear with a percentage that varies between 60% and 80% when velocity varies from v = 3 m/s to v = 5 m/s for different number of UAVs. Moreover, we also monitored the occupancy of altitude levels in RVM, which results from random vertical maneuvering. The available altitude levels were occupied with an average of 1 to 3 UAVs per level.

In contrast to RVM, OVM only allows one UAV to maneuver based on a predefined condition and the occupancy of altitude levels.

Table 4. OVM performance results when v = 3.

N	Detected Conflicts	Handled Conflicts	3D–3D Conflicts	3D–2D Conflicts	2D–2D Conflicts	Maneuvering Energy Consumption
20	9	4	6	3	0	3.26
25	15	6	8	6	1	7.2
30	22	9	12	9	2	11.27
35	29	11	15	12	2	17.03
40	37	13	18	15	5	27.37

shows OVM performance for a different number of UAVs when velocity v = 3 m/s. Similar to SLIDE and RVM, in OVM, conflicts are detected by both conflicted UAVs. However, only one UAV is taking the maneuver. Table 4 shows the conflicts predicted by SLIDE. Our approach focus is on 3D-to-2D and 3D-to-3D conflict resolution. Hence, the results of the OVM algorithm are shown in the total handled conflict column to indicate the total number of maneuvering UAVs that moved to 2D plane. While only one UAV is maneuvering per conflict in ideal situations, number of resolved conflicts is more than half due to the asymmetric conflict explained previously. On the other hand, 3D-to-2D and 2D-to-2D conflicts represents the conflicts introduced in the OVM approach. Consumed energy represents only one conflicted UAV maneuvering in the occupancy-based approach, which is expected to result in less energy consumption than moving both UAVs per conflict. In

Table 5. OVM performance results when v = 5.

N	Detected Conflicts	Handled Conflicts	3D–3D Conflicts	3D–2D Conflicts	2D–2D Conflicts	Maneuvering Energy Consumption
20	15	6	8	5	1	6.77
25	24	9	11	11	2	14.79
30	35	12	17	15	3	25.19
35	48	16	21	21	6	46.81
40	65	20	25	30	10	72.15

, the occupancy-based algorithm experiment is conducted with velocity v = 5 m/s and a different number of UAVs.

In Figure 7, we compare the consumed energy for maneuvering in OVM for different velocity values (v = 3 m/s and 5 m/s) and numbers of UAVs. The energy is affected by the choice of level made by each UAV. In OVM, UAVs make energy-aware maneuvering decisions as they choose to move toward the least congested and nearest level. Hence, it tries to reach the ideal occupancy of levels by balancing the distribution of UAVs over the available altitude levels whenever a conflict is detected. However, balancing the occupancy becomes more challenging when velocity is high and there are a larger no. of UAVs, as the figure and above tables show. We also observed that with a higher no. of UAVs, maneuvering becomes more frequent, and UAVs need to move farther, as the middle levels are permanently occupied by other UAVs. This balancing behavior of OVM resists the fact that nodes tend to move back to the middle levels, which is indicated by the RWM mobility model in [12].

5.4. Performance Comparison of RVM and OVM

In this section, we compare the performance of RVM and OVM. Figure 8 shows the no. of introduced conflicts by RVM and OVM when v = 3 m/s. It is clear that RVM creates more future conflicts than OVM as a result of its random vertical maneuvers. Moreover, Figure 9 shows a significant reduction of energy consumption by OVM compared to RVM. Hence, under the same scenario, OVM consumes less energy than RVM for the same number of UAVs. Finally, in Figure 10, we show the no. of successfully handled conflicts by OVM and RVM.

6. Conclusions

The increased autonomy of UAVs has created significant difficulties in ensuring the safe operation of UAVs in non-segregated airspace. Conflict detection and resolution is an important tool to ensure the safe operation of multiple autonomous UAVs. This work proposes a conflict resolution approach to extend an existing efficient conflict detection algorithm called SLIDE. Our approach includes two possible algorithms for performing vertical maneuvers to resolve conflicts.

We evaluated the performance of our proposed approach through simulation. Several performance measures are considered to assess the performance of the proposed RVM and OVM. The comparison of both algorithms is conducted through three experiments with different numbers of UAVs and velocities. The simulation results show that all 3D–3D and 3D–2D conflicts were resolved for both algorithms RVM and OVM. While RVM is a simple and fast technique to select the level, OVM is more context aware and efficient. Overall, OVM outperforms RVM. There are several reasons for including two CR methods: (1) to test the effectiveness of our assumptions regarding the underlying model and dividing the airspace in multiple altitude levels, (2) to test the impact of fixing the altitude level of conflicted UAVs with different methods, and (3) to test and compare the impact of selecting the alternative levels to maneuver after a conflict is detected. This comparative study may interest many researchers in extending the novel work presented in this paper.

We conclude that CR represents a complex task that requires the implementation of multiple mechanisms to handle different scenarios. Although OVM has an improved performance, from a modeling and design perspective, calculating occupancy can be challenging in a large-scale network with a large number of UAVs that are flying autonomously at high speed. For future work, we plan to simulate our approach for a larger number of UAVs (≅100) and extend it to include a resolution of 2D–2D conflicts. We also plan to design an energy-aware occupancy-based algorithm to improve its energy efficiency.