Systematic Review of Multi-Objective UAV Swarm Mission Planning Systems from Regulatory Perspective

Abstract

Advancements in Unmanned Aerial Vehicle (UAV) technologies have increased exponentially in recent years, with UAV swarm being a key area of interest. UAV swarm overcomes the energy reserve, payload, and single-objective limitations of single UAVs, enabling broader mission scopes. Despite these advantages, UAV swarm has yet to see widespread application within global industry. A leading factor hindering swarm application within industry is the divide that currently exists between the functional capacity of modern UAV swarm systems and the functionality required by legislation. This paper investigates this divide through an overview of global legislative practice, contextualized via a case study of Australia’s UAV regulatory environment. The overview highlighted legislative objectives that coincided with open challenges in the UAV swarm literature. These objectives were then formulated into analysis criteria that assessed whether systems presented sufficient functionality to address legislative concern. A systematic review methodology was used to apply analysis criteria to multi-objective UAV swarm mission planning systems. Analysis focused on multi-objective mission planning systems due to their role in defining the functional capacity of UAV swarms within complex real-world operational environments. This, alongside the popularity of these systems within the modern literature, makes them ideal candidates for defining new enabling technologies that could address the identified areas of weakness. The results of this review highlighted several legislative considerations that remain under-addressed by existing technologies. These findings guided the proposal of enabling technologies to bridge the divide between functional capacity and legislative concern.

1. Introduction

Drones, or Unmanned Aerial Vehicles (UAVs) have seen a groundswell in application following advancements in UAV hardware, and UAV control systems. Their popularity is owed to the unique operational advantages intrinsic to UAV systems, including mobility, versatility, ease of deployment, and low running costs [1]. With these advantages, single-UAV systems have seen growth in a wide array of industries [2] such as agriculture [3], construction [4], entertainment, surveying [5], and emergency services [6]. However, as the service envelope of single-UAV platforms continues to increase operators have encountered several challenges, namely, restricted payload sizing, limited available energy reserves, limited sensing ranges, and restriction to singular concurrent objectives [7,8]. Drawing inspiration from nature, UAV swarm systems were developed to make-up for the deficiencies of single-UAV systems through the collaborative effort of multiple individual agents [2,9,10]. By facilitating inter-drone collaboration UAV swarm operators enjoyed several other advantages including, major reductions in operation time, increased operational versatility and increased operational robustness [7]. These advantages have been compounded through the continued development of a plethora of new enabling swarm technologies, one such technology of note to this paper is multi-objective mission planning systems which have helped UAV swarm systems to address multiple concurrent objectives [11,12]. However, despite these advantages, UAV swarm systems have failed to see widespread application within industry, only seeing fringe use in non-military setting. Modern review literature explored possible causes for this disparity, identifying regulatory restriction as a major factor limiting UAV swarm application into industry fields that already employ single-UAV systems [13,14,15]. The purpose of this paper is to further the investigation into the regulatory restriction of UAV swarms by directly investigating what common legislative considerations are unaddressed by modern UAV swarm systems. This paper aims to achieve this through the completion of a novel systematic review of UAV swarm multi-objective mission planning systems through the lens of international legislation.

Although there is this groundswell in UAV swarm technical development, operational considerations such as real-time mission observability, multi-user control, link loss mitigation, and integrated dynamic geofencing, are not yet comprehensively addressed, particularly when considered from a regulatory perspective. By identifying and better defining these unaddressed requirements this paper seeks to direct future UAV swarm research onto key areas of weakness, which if addressed, will help foster UAV swarm application into a range of new markets. Thus, the core contributions of this paper are as follows: (1) application of a novel systematic framework that focuses on a unique research perspective, (2) identification of key discrepancies between UAV swarm functional capacity and legislative concern that are limiting UAV swarm application, and (3) proposal of targeted research directions aimed at bridging this divide.

Multi-objective mission planning systems were selected as the technical focus for this review because of their role in defining the core functional capabilities of applied swarm systems [16], and popularity as a modern control architecture [17]. Doubtlessly, mission planning systems are not the only mechanistic component of UAV swarm that foster unaddressed legislative requirements but to manage the scope of this review they are analyzed in isolation. The analysis criteria employed during the review process are defined using supporting legislative instruments from Joint Authority’s for Rule making for Unmanned Systems (JARUS), an international body constituted of members from regulatory bodies world-wide [18]. To support the use of JARUS’s documentation a case study of the Australian drone regulatory process is preformed to both articulate how JARUS’s instruments are applied within regulatory bodies and to demonstrate how they align with current industry pressures and potential areas of regulatory change. The developed criteria are defined to directly evaluate whether a reviewed article addresses the functional requirements detailed within legislation. Given the extensive nature of UAV regulation [19], it would be infeasible to address all legislative requirements within a single paper. Thus, to further manage the scope of the review key requirements of interest must be identified. This is performed using a cross-referencing strategy, which compared the legislative functional considerations with open challenges defined within existing UAV swarm review papers.

This paper consists of six sections. Section 1 describes the supporting background information utilized by the later sections of the paper. This includes information pertaining to the current state of global drone legislation, current state of UAV swarm industry and the current state of UAV swarm technologies. Section 2 describes the specific contributions of this paper and compares these contributions to other modern UAV swarm reviews. Section 3 describes the methodology of the systematic review. This includes the article selection process, the justification of the key legislative considerations of interest, and the definition of the analysis criterion. Section 4 then details the results of the systematic review. Section 5 then provides a detailed discussion surrounding what aspects of legislation remain unaddressed by the capabilities of UAV swarm systems. Section 6 proposes potential enabling technologies that could address the identified areas of weakness.

2. Background

2.1. State of UAV Swarm Legislation

Globally UAV legislation is seen as a contentious issue, being cited as a key factor limiting UAV swarm application within industry [20]. This discussion is relevant to UAV regulators worldwide and is keenly felt across local industries in multiple countries, for examples, Australia [15,21,22] and the United States of America [23]. This issue is the product of two conflicting perspectives, highlighted in

Table 1. Common contentions surrounding UAV legislation.

Supporting Arguments	Opposing Arguments
Safety Assurance: Current legislation has maintained safe airspaces for civilians and workers alike [24]	Stifles Innovation: Overbearing regulations prevent rapid testing and novel drone applications [15,20]
Accountability and Traceability: By enforcing operator registration and on-site operation accountability is assured [25,26]	Overreliance on Human Operators: One pilot one drone rulings directly oppose swarm operation.
Security and Privacy Protection: Prevents misuse of UAV systems for illegal means [19]	Inconsistent Standards: Divergent rule sets among regulators especially in regard to privacy and flight pathing [22]

. Swarm regulators staunchly support current legislative practice, a stance bolstered by the safe-work trends seen in major markets worldwide. Whilst swarm applicators criticize current practice citing it as restrictive and overbearing. Notionally there are only two plausible resolutions for this issue. Either legislation is amended to reduce the requirements placed upon UAV swarm operators, or new enabling technologies are developed that satisfy current requirements. Since innovators cannot influence legislative change directly, emphasis should be placed on developing compliant technologies to strengthen their case for legislative reform. To achieve this the legislative structures used to regulate UAV systems must first be understood.

Defining UAV regulation from a global perspective is a difficult task as there are no internationally recognized legislative instruments that address UAV systems [19,27]. As such, most nations have implemented unique regulatory approaches [19], as shown in Figure 1. This fragmentation inhibits how UAV legislation can be investigated from a global perspective. However, it has also prompted the formation of international advisory bodies that promote cohesive and progressive material to support UAV regulators with policy reformation and education [28]. JARUS is one such advisory body which is comprising representatives from over 67 countries and 69 member organizations [18]. The works of JARUS draw upon the efforts of these member organizations to define technical, safety and operation requirements that are inclusive of the clauses posed in existing regulations [28]. Because of this extensive body of support, the works of JARUS have already seen adoption in multiple countries, including one of the world’s longest standing aviation regulators, Australia’s Civil Aviation Safety Authority (CASA) [29]. Thus, JARUS’s works can serve as a body of requirements representative of the standards held by UAV regulators globally in absence of internationally recognized legislative instruments.

JARUS provides the internationally recognized Specific Operations Risk Assessment (SORA) method, which details a range of operational requirements that need to be addressed to ensure the safe operation of any UAV system [30]. SORA is an analytical based risk assessment system that has seen widespread application globally [31], as detailed in

Table 2. Interrelation between JARUS and major UAV regulator bodies [18,28].

Regulator	Country of Origin	JARUS Member	Legislative Implementation
CASA	Australia	Active	Early legislative adopter of JARUS, directly incorporated SORA into legislative documentation.
FAA	United States	Active	Actively assisted in the development of the SORA documentation. Advertises its use throughout its legislative processes.
CAAC	China	Active	Participates in JARUS discussions but internal regulations are independently referenced.
EASA	European Union	Active	A leading member of JARUS, EASA regulations closely align with JARUS instruments including the SORA.

. It includes a core document [32], several supporting annexes, and other supporting materials. This system separates UAV operations into three categories based on the operation requirements, system elements and the assumed risks associated with a given application [33]. Swarm applications are addressed under category B. Category B systems must adhere to the considerations presented within the SORA and are only permitted upon receipt of authorization from the relevant authority, unlike category A systems which can be undertaken without prior permission. This categorization is consistent with the regulatory approaches used by several countries including Australia, South Korea, India and the United States of America [19].

The SORA documentation requires operators to consider a broad spectrum of operational states and circumstances. Its analytical system centers around three categories of harm, damage to ground personnel, damage to aerial objects and damage to critical infrastructure [32]. This system proposes a logical process that defines, analyzes and controls the risks associated with each of the three categories of harm. Throughout this process the SORA presents an extensive list of functional considerations for UAV systems [32], A list extended by the SORA CONOPS [34]. Whilst the SORA’s core tenets are defined for single drone operations, as per category B’s definition [33], the general CONOP structure and SORA considerations are applicable for swarm applications. Thus, SORA presents a globally relevant reference for innovators developing compliant technologies.

2.2. Australian Regulatory Enviroment

To contextualize this investigation, a case study of the Australian regulatory environment was conducted to demonstrate how current legislation is restricting UAV swarm application. The work of [15] states that the application of drone networks (swarm systems) within Australian industry is infeasible due to regulator inflexibility and rigorous regulatory concern. Whilst [15] focuses on drone medical delivery, the presented criticism of the regulator (CASA) extends beyond this domain as the jurisdiction of CASA’s UAV regulations are application-agnostic. The works of [21,22] reinforce this criticism, advising applicators to be aware of the challenges imposed by the current regulatory system. The restriction of UAV swarm within Australia is also clear from an industry perspective. UAV swarms have only seen fringe commercial use in Australian industries, namely entertainment [35], and agriculture [36]. In contrast, single-UAV systems have seen a sharp uptake in application [35], spanning across a range of major Australian industries such as mining, agriculture, construction, aerial surveying, and search and rescue [13].

The contention surrounding CASA’s legislation is centered around two legislative instruments, CASR part 101 “unmanned aircraft and rockets” [37] and the Manual of Standards [25]. These documents are condensed into CASA’s drone safety rules [38] which define the standard operating conditions for all recreational or commercial flights. The first drone safety rule actualizes the distinct regulatory difference between single-UAV and UAV swarm flights, stating “you must only fly one drone at a time” [38]. This rule conflicts with a core tenet of UAV swarm systems, forcing all such systems to acquire pre-flight approvals (“Flight Authorizations”) [26] which have hindered their adoption into industry [15]. The primary goal of flight authorizations is to ensure that drone operators have controlled the additional risks inherent in applications that violate the drone safety rules. This process uses an application and approval framework that enables CASA to check that all reasonably foreseeable risks have been controlled prior to conducting the flight [26]. Applications require the completion of an approvals form (form varies depending on application, beyond line-of-sight approvals require [39], swarm systems require [40]), supporting documentation and a risk assessment [26]. CASA assesses this flight approval approach using the general forward regulatory cycle [41] shown in Figure 2.

Returning to the findings of [15,21,22], CASA is aware of the criticisms of the current structure thanks in part to its consultation forum and post-implementation review process [42]. In response, CASA published a roadmap [43] alongside a supporting consultation [44], which detailed regulatory reform under the “Forward Regulatory Program” [45], and an increased emphasis on supporting educational materials [43]. Whilst these changes aim to reduce red tape for the UAV industry [46], they reaffirm CASAs staunch stance on maintaining safe operations through rigorous risk management-based approvals processes [43,44,45,47]. In reference to UAV swarm, the work in [36] demonstrates how CASA is actively advertising the current approvals process as an industry ready method for safely approving commercial swarm operations. To further cement the existing process CASA has provided educational material focused on removing ambiguity surrounding the technical and operational requirements for flight approval [43]. SORA is routinely referenced throughout these changes a model framework suitable for Australian operators. This includes references within existing flight authorizations documentation such as [48] and CASA’s road map [43]. Challenges arise as several of the functional considerations presented within SORA remain unaddressed by current UAV swarm technologies as will be discussed in Section 4.2.2.

2.3. State of UAV Swarm Technology

UAV Swarm systems are traditionally defined to consist of four major mechanistic components with each having its own unique role [8,49,50]. These four components are, task assignment, path planning, formation control and a communication control [8,49,50]. Task assignment determines the next action for each individual UAV given a mission context, and as such it defines the functional capability of the UAV swarm [51,52]. The path planning component directs individual UAVs toward the completion of said action given a mission environment, and as such it defines the type of environment the swarm can navigate [52]. Formation control maintains drone spacing to avoid intra-swarm collisions [49]. Finally, the communication control maintains intra and inter swarm data links [7,53]. The design of these swarm components has been traditionally influenced by four independent characteristics that change the way in which the swarm operates as a holistic system. The four characteristics are, the command architecture [54,55,56], the homogeneity of the swarm [57], the operation context [58,59] and the size of the swarm [60]. The command architecture defines the way the swarm deliberates and disseminates information. There are two main command architectures, centralized, where all UAVs are controlled from a single node and decentralized, where all UAVs contribute to the decision-making process. The homogeneity of the swarm refers to the type of drones that make up the swarm. Operation context considers whether mission objectives are fixed (offline) or evolving (online), and whether the mission environment is static or dynamic. The size of the swarm refers to the number of drones within the swarm.

Modern literature optimizes UAV swarm systems by coupling the task assignment and path planning components into a single mission planning module to incorporate task locations into the task assignment process [51,52,58]. Given this coupling the mission planning component must now holistically consider the function of the swarm accounting for both its mission objectives and mission environment [51,52]. This concept is shown in Figure 3, which displays a centralized mission planning system communicating trajectory paths to four UAVs, whilst accounting for optimal obstacle avoidance and optimized task assignments. This holistic consideration of swarm function makes mission planning systems a pivotal enabling technology, especially when considering the addition of new UAV swarm functions. Consequently, the design of mission planning algorithms is heavily influenced by the characteristics and functional requirements of the proposed swarm. These factors influence algorithm selection and may require additional functions. The first factor is the operation context, online systems require both task pre-assignment and task re-assignment functions [52], whilst offline systems only require task pre-assignment [52]. The purpose of the re-assignment function is to adapt to new information (or tasks) by altering individual UAVs current task assignments during run-time. The second influential characteristic is the command architecture, centralized structures handle task assignment within a central node, however in decentralized structures a dissemination function is required to share information. The purpose of this dissemination function is to handle task competition by allowing the collective to exhort influence into each drone’s task assignment process without requiring a centralized node to arbitrate. These two additional functions are typically built into the task assignment subcomponent.

Task assignment algorithms used within mission planning fall into three main categories, each with unique advantages and disadvantages. These categories are mathematical approaches [51,52], swarm intelligence approaches [9,50,52] and machine learning approaches [51,62]. Mathematical approaches are the traditional method used to solve task assignment problems [52], whereby all candidate solutions are compared directly through some mathematical method [51]. Advantages of this approach are high accuracy and high adaptability [51]. However, as agent and task counts increase these approaches require exponentially more computation, which can slow convergence times, and limit application in distributed systems where onboard computational resources may be limited [51,52]. Modern mathematical approaches have adjusted these drawbacks by implementing heuristics (to solve the convergence and scaling problems) and negotiation algorithms (to solve the reliance on centralized nodes) [51]. Swarm intelligence approaches use an iterative random search process whereby some heuristic information is generated and shared to inform subsequent search stages [50]. Advantages of this approach are scalability, applicability in decentralized architectures and robust operation; however, they can convergence to locally optimum solutions [52] and have slow convergence speeds [9,50]. A myriad of improvements have been made to address the disadvantages sited above; however, in addressing these disadvantages, the improved algorithms are often hyper-specialized and cannot be transferred to adjacent problems [9,50]. Machine learning approaches use a trained model to preform task selection and delegation [51], drawing upon a diverse range of machine reinforcement or federated learning approaches applied in other fields of research [51]. Their advantages include excellent convergence speeds, global optimality and robust performance in dynamic or uncertain environments [51,62]. However, they require extensive training and can suffer from stability issues [51].

A sub-genre of mission planning systems of particular interest to this review is multi-objective mission planning. Multi-objective mission planning extends the functionality of traditional mission planning problems by considering multiple concurrent and competing objectives. An extension essential to modern UAV swarm application within industry. This sub-problem was first adopted into the context of UAV swarm back in 2004 through the works of [63] and has seen an exponential increase in publication in recent years. Within modern literature this problem is interpreted in two ways, the first is the completion of a single overarching task whilst optimizing multiple competing objectives [64,65,66] and the second is the completion of multiple competing unrelated tasks [67,68,69].

3. Contributions

Commensurate with the popularity of UAV swarm research, a multitude of review papers have been published on the subject. This paper’s primary contribution to this body of research is the results and future research directions found using this review’s unique research perspective. This paper’s perspective was adapted from two areas of existing research; these areas are surveys addressing UAV swarm mission planning systems and UAV swarm legislation.

Current reviews regarding UAV swarm mission planning primarily focus on analyzing the functional capacity of the analyzed systems mechanistic components directly, proposing deficiencies found through comparative analysis of the algorithms/systems themselves. Recent publications include: The review [11] published in January 2025, which focuses on modern task assignment specifically those designed for dynamic environments. Notably this paper raises the importance of continued research on multi-objective algorithms. The review [16] published in February 2024, focuses on the essential technologies enabling collaborative mission planning in UAV swarms. It discusses advancements in communication protocols, coordination algorithms, and decision-making processes that facilitate effective swarm operations. The work [12] published in July 2024, provides a comprehensive review examining the application of constrained multi-objective evolutionary algorithms within UAV swarm mission planning systems. It discusses the challenges of optimizing multiple conflicting objectives under various constraints and evaluates the effectiveness of different constraint-handling techniques. The study [51] published in February 2023, provides an updated overview of mission planning problems in UAV swarm systems by comparing various mathematical methods, heuristic algorithms, negotiation algorithms, and neural networks. In contrast to these three examples, other reviews implement an external set of requirements to analyze the algorithms from a particular perspective, notionally a specific industry application is used to inform this process. The survey [70] published in August 2024, examines the diverse applications of UAV swarms across sectors such as reconnaissance, search and rescue, and agriculture. It delves into mission planning strategies, highlighting the challenges and opportunities in deploying UAV swarms for various objectives. This is mirrored in several other reviews [21,54,71], which address more industry applications such as mining and agriculture.

Current reviews regarding the second area of interest of UAV swarm legislation are primarily directed toward analyzing the legislation’s impact on a specific application or toward ratifying the legislation itself. Whilst this area is less popular than the UAV swarm technologies topic, it has still seen a number of recent publications. The research [19] published in 2019, is an example of the later approach which is commonly taken when discussing UAV swarm legislation. This paper lists and analyzes the legislative processes used by a range of countries and makes no direct considerations for UAV technologies. The work in [31] published in 2025 extends previous legislative investigations by focusing specifically on the safety assessment techniques used to evaluate UAV systems. Again, this paper is an example of the former approach focusing on the analysis methods rather than specific mission planning systems. The research [15] published in 2023, is an example of the former approach and focuses on the restrictive nature of current UAV swarm legislation and the impact this has on the medical delivery field. In contrast to the articles presented above, the perspective of this review is unique as it uses the technical requirements proposed within legislation to analyze the current state of multi-objective mission planning systems. To the best of the authors of this paper’s knowledge this review is the first to adopt this perspective.

4. Method

4.1. Article Selection

4.1.1. Selection Overview

This paper employs a systematic process to identify relevant articles for inclusion within the review. The protocol used to guide this process is adapted from [72], and consists of three review stages, identification, screening and eligibility assessment. The first stage, identification, accessed four candidate databases, ACM Digital, IEEE, SCOPUS and Web of Science, using the following search string: (“UAV Swarm” or “UAS Swarm” or “RPAS Swarm” or “Multi-UAV” or “Multi-UAS” or “Unmanned Aerial Vehicle Swarm”) AND (“Multi-Objective” or “Multiple Objectives” or “Multi-Task” or “Multiple Tasks”). Validity of the search string was confirmed via keyword analysis of the article base performed using VOSViewer, results shown in Figure 4. This keyword analysis detected no other major keywords common amongst the candidate articles, indicating that the search string was not missing any major terms. Article review period was from 2007 to the end of quarter one 2025.

Applying the search string and search range to the four candidate databases, an initial pool of 282 articles was retrieved. The second stage, screening, removed duplicate, inaccessible and irrelevant articles from the initial pool, following this stage 135 articles remained. The final stage, eligibility assessment, uses the selection criteria and quality assessment checklist to evaluate whether an article should be included within the final review pool. The selection criteria and quality assessment checklist are detailed in Section 4.1.2 and Section 4.1.3, respectively. Following the eligibility assessment 55 articles remained for direct analysis. This analysis process is summarized in Figure 5.

4.1.2. Selection Criteria

The applied selection criteria are split into two sub sections, inclusion criteria and exclusion criteria. The inclusion criteria are as follows: Original research articles are to be included within the review if they present, (a) a design of a full or a sub-component, (b) an application, (c) an evaluation, (d) an algorithm or (e) any other aspect, of a multi-objective mission planning system. The exclusion criteria detail exceptions to the inclusion criteria, excluding articles if: (a) the article is published as a review, survey, thesis, or pre-publication, or (b) articles published on non-UAV multi-objective systems, (i) not considering task assignment, (ii) considering mission planning systems but only for single-UAV systems, or (ii) considering multi-objective control for only single concurrent task missions.

4.1.3. Quality Assessment

The applied quality assessment checklist is adapted from [72], which adjusted the widely used qualitative checklist presented in [73] to a technology domain using the suggestions proposed in [74]. The purpose of this assessment is to remove unverifiable or unsubstantiated articles from the candidate article pool. The applied checklist focused on using six checks directed toward verifying the validity of each article’s presented methodology, unlike the eight broader checks used in [72]. The applied checklist addressed the following topics: (a) paper context, (b) problem definition, (c) method constraints, (d) evidence of method validation, (e) credibility of validation, and (f) study reflectivity. Like [72], articles receive a yes (2), partial (1) or no (0) score for each topic. If the total exceeds 5 the article is cleared for inclusion within the review.

4.2. Article Analysis

4.2.1. Criterion Overview

Using the globally recognized SORA framework as a legislative baseline, this review cross-references open challenges in UAV swarm literature with SORA defined functional considerations and contextualizes them using legislative references from regulators such as CASA, FAA and CAAC. This ensures the analysis remains internationally relevant while addressing real-world implementation barriers observed in national contexts. Due to the number and diversity of the considerations presented within SORA it is infeasible to assess each function directly during the review process. To manage this expansive scope without counterfeiting the core objective of the review, open challenges referenced in existing literature were cross referenced with these considerations to identify key areas of interest. These identified areas were then formulated into a set of analysis criteria, with each criterion being comprising multiple single-mark metrics. These metrics were defined to qualitatively assess if a reviewed system presented technical mechanisms capable of satisfying the legislative requirements relevant to that analysis criterion. Each criterion’s final score was calculated from these metrics and displayed as a percentage of awarded marks over the rubrics maximum score for easy of display. This formulation method ensures that this review builds upon existing UAV swarm literature and efficiently achieves its purpose. In total four analysis criterion were derived to assess the following four areas of weakness: UAS Instruction from multiple external observers, support for dynamic no-fly-zone geofencing, command and control link loss stability and UAV swarm observability. The following sections discuss the individual criteria in detail, describing the primary legislative and SORA references, supporting open challenges that lead to their selection and finally the applied marking rubric.

4.2.2. Criterion One—Instruction from Multiple External Observers

The first criterion assessed how an analyzed mission planning system addressed heterogenous commands issued from distributed users. This is an essential function for the safe operation of any industry system, especially when considering functions such as emergency stops, objective adjustments or new commands. From a global perspective this criterion references CONOPS section A.2.3.2 (3) [34] which directly considers how UAS systems should respond to instructions from multiple concurrent and unique users such as air traffic controllers, visual observers, and other crew members. These considerations appear in local legislation in both the United States and Australia, which reference UAV systems being capable of accepting commands from sources external to the primary operator such as air traffic controllers and emergency services. Two components are required to perform this function, a human computer interface capable of interpreting commands from distributed sources and a mission planning system capable of adjusting swarm function to distribute and assimilate multiple commands simultaneously. Out of these two components, the mission planning system is selected as the key technical interest for analysis under this criterion. This was performed to reflect the views presented in modern UAV swarm surveys, which posit the existence of multiple open challenges adjacent to the role of the mission planning component. The work in [11] cites open challenges regarding both multi-objective optimization and real-time adaptability which are two vital functions required by the first core component. Subsequently the reviews, ref. [2] (2022), ref. [50] (2019) and [20] (2022) supported this discussion all citing “Distributed UAV architectures addressing multiple objectives” (or similar) as a current open challenge.

This criterion’s rubric focused on assessing if the analyzed system presented the swarm characteristics necessary to achieve distributed multi-user control. To achieve this, this rubric consists of four single mark metrics assessing command architecture (1A), operation context mission information (1B), task heterogeneity (1C) and swarm heterogeneity (1D).

• 1A. This metric assessed the command architecture by qualitatively characterizing the control architecture used. One mark was awarded if a decentralized architecture was used during operation. That is, if a centralized method was used during swarm training, but a fully distributed architecture was used during operation, one mark would still be awarded. Half marks were awarded if the system demonstrated hierarchical control and zero marks awarded if the system relied on a central control node.
• 1B. This metric assessed operation context by qualitatively identifying whether online or offline control was used. One mark was awarded if an online task assignment method was used. Zero marks were awarded if an offline task assignment method was used, that is an omniscient knowledge of available tasks was required prior to mission commencement.
• 1C. This mark assessed task heterogeneity by classifying uniqueness of tasks considered during the task assignment process. If each task had more than one unique characteristic that directly influenced the applied task assignment algorithm one mark was awarded else, zero marks were awarded. For example, if tasks were defined to consist of a unique location and a unique name but the given names had no influence on task assignment then zero marks were given.
• 1D. This metric assessed swarm heterogeneity by classifying the uniqueness of individual UAV agents considered during the task assignment process. If the UAVs had more than one unique characteristic that directly influenced the applied task assignment algorithm, one mark was awarded else, zero marks were awarded.

4.2.3. Criterion Two—Dynamic No-Fly-Zone Support

The second criterion assessed if analyzed mission planning systems could accept, change and re-optimize mission plans to account for dynamic no-fly zones. Adjusting and enforcing no-fly zones is essential during UAV swarm operations to ensure robust avoidance of ground personnel and critical infrastructure. From a global perspective this criterion references CONOPS section A.2.4 (a) [34] which considers the importance of geofencing functions capable of both avoiding and or confining the swarm to specific areas. These considerations are supported by regulators in the United States, China, Japan, and Australia, all of which require, in some capacity, no-fly zones to be maintained around people and essential infrastructure. These no-fly zones are often requested to have specific geometries that must be maintained around a potentially dynamic system element such as a person [38]. Whilst UAV swarm object avoidance systems have been heavily researched, there are multiple open challenges relating to the extension of these systems to facilitate dynamic online geofencing, especially in relation to the interlinking between path planning avoidance and task re-assignment. Reviews such as [75] site that “dynamic obstacles (and no-fly zones) can be challenging to detect and avoid in real-time” especially in decentralized systems. The work in [76] reinforces the above, emphasizing the potential limitations of current task assignment and path planning methods when considering dynamic and unpredictable environments.

This criterion’s rubric focused on how zone/obstacle avoidance methods were integrated within the analyzed system. This assessment primarily addressed the command architecture and operation context characteristics, aiming to analyze if a given system was able to accept new no-fly zone requests, adapt to dynamic or moving regions and effectively adjust operation accordingly. To achieve this, this rubric comprised four single mark metrics that assessed environment obstacle type (2A), obstacle information architecture (2B), obstacle detection (2C), and obstacle submission (2D).

• 2A. This metric assessed the obstacle types considered within the mission planning system using qualitative classification. One mark was awarded if dynamic obstacles were explicitly considered, i.e., obstacles in motion, not dynamic obstacle detection. Half marks were awarded if only static obstacles, static no-fly zones or only inter-drone collisions were considered. Zero marks were awarded if no avoidance method was included.
• 2B. This metric assessed how obstacles were handled within the command architecture by qualitative analyzing how the swarm stored obstacle information. One mark was awarded if the obstacle information was distributed amongst agents. A half mark was awarded if obstacle information was stored independently by each agent. Zero marks were awarded if the swarm relied on omniscient knowledge of obstacle locations or had no obstacle information (did not consider obstacle avoidance).
• 2C. This metric assessed obstacle detection by observing if an online method was used to locally detect new obstacles. This metric extended the investigation into the operation context characteristic, specifically focusing on obstacle information instead of task information. One mark was awarded if an online obstacle detection system was presented. Zero marks were awarded if no obstacle detection method was presented.
• 2D. This metric assessed obstacle submission by observing if a method was presented to accept new obstacle information from an external source. One mark was awarded if an online submission method was presented. Half a mark was awarded if an offline method was presented and zero marks were awarded if obstacle submission was not considered.

4.2.4. Criterion Three—Command and Control Link Loss Stability

The third criterion assessed if analyzed mission planning systems implemented safeguards against both intra and inter swarm communication loss. Given that communication stability cannot be assumed perfect for every given industry application, operators must be able to ensure the continued safe operation of the swarm even in events when communication lines are lost. From a global perspective this criterion references CONOPS sections A.2.6, A.2.7 and A.2.8 [34] which all specify functions that expand upon the command-and-control link, and specifically command-and-control link loss robustness. Control link loss considerations are also directly maintained by a number of regulators including the FAA, EASA and CASA, all of which call upon operators to identify or define procedures for such events. Ref. [76] presents open challenges regarding reliable communication, emphasizing the importance of optimizing UAV control systems to minimize communication loading. Supporting this conclusion the reviews [20,49,51,58,77] all present similar open challenges related to this topic.

This criterion’s rubric focused on the communication considerations made by the analyzed system. This investigation extends the command architecture characteristics by assessing how well the architecture’s communication network is defined and what considerations have been made regarding network failure and repair. To reflect this, this rubric consisted of four single mark metrics: definition of networking constraints (3A), definition of network model (3B), considerations for failure detection (3C), and considerations for network failure mitigation (3D).

• 3A. This metric assessed network constraints by analyzing three network performance metrics, bandwidth, latency and range. One mark was awarded if constrained definitions were provided to limit all three of these key network performance indicators. Constraints could either be imposed directly using numerical limits or indirectly through network component definition. Half marks were awarded if only subsets of the factors were constrained, and zero marks were awarded if none of the factors were considered.
• 3B. This metric assessed the networking method by qualitatively classifying the complexity of the method used. One mark was awarded if a specific networking structure was provided, this structure could relate to either inter-swarm or intra-swarm communication providing it handled the communications required for UAV tasking. Half marks were awarded if a numerical representation was used, i.e., graph structures or adjacency matrices. Zero marks were awarded if no discussion of the networking method was provided.
• 3C. This metric assessed communication failure detection by analyzing what possible failures had been acknowledged and if so, what methods had been proposed to detect the referenced failure types. One mark was awarded if one or more failures were considered and detection methods proposed. Half marks were awarded if failure cases had been acknowledged but not detection method was proposed. Zero marks were awarded if communication failure was not considered.
• 3D. This metric assessed failure mitigation by investigating if communication recovery or link loss damage mitigation method were proposed. One mark was awarded a mitigation method was implemented and its functionality demonstrated during algorithm validation. Half mark was awarded if a mitigation method was proposed by not implemented, or if speculation of possible mitigation methods was offered. Zero marks were awarded if no communication failure mitigation or recovery methods were not discussed.

4.2.5. Criterion Four—UAV Swarm Observability

The fourth criterion assessed if analyzed mission planning systems implemented discrete return methods for critical mission data to the user. Whilst this is not directly linked to the control of UAV swarm systems, it is vitally important within industry applications as no system can operate without oversight. From a global perspective this criterion references CONOPS section A.2.3.5 (2) [34] which considers how operators can determine critical flight parameters for individual UAVs within the swarm. This combines with A.2.3.5 (3) [34] which discusses how this information can be conveyed to the operator and to other system users such as air traffic controllers. This requirement is not commonly included in legislation directly; however, both EASA and CASA routinely referenced similar considerations as part of the operational risk assessments they require for UAV system approval. The works of [78] clarify that this functionality should be considered under the purview of mission planning systems, highlighting the importance of such functionality to inform users and allow for real-time decision-making. Whilst this problem is largely solved for centralized swarm architectures it is still sited as an open challenge especially for de-centralized swarm architectures. The generation and distribution of this information fall under the purview of mission planning systems and is mentioned as an open challenge in [50]. The presentation of this information is cited as an open challenge in [78].

This criterion’s rubric aimed to assess how the analyzed system considered performance validation within the contexts of both algorithm validation and swarm observability. To achieve this, this rubric extended the characteristic of operation context to consider how information is relayed during validation and what information is used to validate swarm performance. This rubric was comprising four single mark metrics which assessed standard operation evaluation method (4A), performance data observation method (4B), unsolicited event evaluation method (4C), and unsolicited event data observation method (4D).

• 4A. This metric assessed the evaluation method used to validate mission planning performance under standard operating conditions using a qualitive classification. One mark was awarded if a simulation environment was used to holistically validate swarm performance within a specific operating scenario. Half marks were awarded if numerical analysis was used to validate swarm optimality. Zero marks were awarded if either validation was not preformed or if the validation method was not detailed.
• 4B. This metric assessed the method used to generate or access swarm performance data during algorithm validation. One mark was awarded if the performance data was observed by the swarm’s mission planning system, and returned to the evaluation platform through a pre-defined data channel. Zero marks were awarded if an omniscient viewpoint was used to ascertain swarm performance data, that is, the validation environment directly accessed UAV location/task assignment information to evaluate swarm performance.
• 4C. This metric assessed the degree to which unsolicited events had been modeled within the mission planning evaluation process. One mark was awarded if swarm validation included unplanned failures such as task failure, communication failure or drone failure. Half marks were awarded if the validation environment allowed for online task or obstacle submission to the swarm. Zero marks were awarded if an offline validation approach was used.
• 4D. This metric assessed how performance/failure data associated with unsolicited events was handled. One mark was awarded if the mission planning system observed and generated failure data directly, relaying this information to environment for response evaluation. Zero marks were awarded if the swarm had no considerations for reporting unsolicited events, and as such, an omniscient viewpoint was used to validate the swarm’s response to said events.

5. Results

5.1. Summary of Findings

The application of the analysis criteria to the candidate articles emphasized several key areas of the divide between legislation and the functional capacity of modern UAV swarm mission planning systems. This analysis showed a continued lack of conformance with no article scoring full marks against all four criteria. However, a key finding of this analysis was demonstrated by the trends observed between the article sets which showed a general movement toward legislative conformity. As shown in Figure 6, this trend was observed across all four of the criteria. The second key finding was demonstrated through the specific rubric results which identified specific areas of weakness preventing legislative conformity. This was supported by the direct analysis of the highest preforming articles, discussed in Section 6.

Analysis driven by criterion one’s rubric showed that most papers only conformed to one or two of C1 metrics, with an average score of 40.91% (results in Appendix A). Modern articles scored higher on average due to research trends that favored hybrid/decentralized architectures and increased problem complexity which promoted swarm heterogeneity. This resulted in the 2025 articles scoring an average of 56.25% and the 2024 articles averaging 42.5%. Another key trend is the grouping of 1A, 1B scores and 1C, 1D scores. Criterion two had an average score of 11.82% highlighting the limited integration of obstacle avoidant path planning methods within modern multi-objective mission planning systems. These findings emphasized the limitations of loosely coupled mission planning subcomponents (results in Appendix B). Modern articles showed minor improvements, with 2025 publications averaging 20.83% indicating that there is limited pressure to in-corporate more complex path planning models into multi-objective mission control systems. Criterion three’s analysis showed limited consideration toward communication modeling, stability analysis and failure rectification, with an average score of 33.86% (results in Appendix C). This criterion also highlighted a promising trend in recent research with 2025 articles scoring 56.25%, and 2024 articles scoring 39.17%. Criterion four had an average score of 27.5%, ode to a limited use of mission critical data observation and generation within the candidate article operation scenarios (results in Appendix D). Furthermore, whilst there was an improvement in 2025 scores (41.7%) over previous years, this increase was not part of an observed trend.

5.2. Article Results

The individual article results are presented in the following four tables. These tables group candidate articles by publication year to further highlight the trends discussed in Section 5.1. These groupings are as follows:

Table 3. 2025 Quarter One Articles.

Ref.	Pub. Year	Developed Algorithm	C1	C2	C3	C4
[79]	2025	BCIA	50%	0%	0%	12.5%
[80]	2025	TDC-MOPSO	87.5%	0%	62.5%	75%
[81]	2025	Improved CNP	87.5%	25%	50%	12.5%
[82]	2025	MT-MDP	37.5%	87.5%	75%	50%
[83]	2025	RCEA	50%	12.5%	75%	75%
[84]	2025	GDMTD3	25%	0%	75%	25%

presents articles published in quarter one of 2025,

Table 4. 2024 Articles.

Ref.	Pub. Year	Developed Algorithm	C1	C2	C3	C4
[85]	2024	p-NSGA-IImv	25%	0%	12.5%	25%
[86]	2024	Modified Hierarchical Auctioning Algorithm	50%	25%	25%	25%
[87]	2024	DCTAEA	25%	0%	25%	12.5%
[88]	2024	CSS-MPCCC	25%	0%	62.5%	12.5%
[89]	2024	Adapted Market-Based Algorithm	50%	0%	75%	12.5%
[90]	2024	DDQN coupled MORL	50%	75%	50%	50%
[91]	2024	MPMFG and DTFSSMFG	0%	25%	0%	12.5%
[92]	2024	Red Fox	50%	0%	0%	12.5%
[93]	2024	MOEA	50%	25%	75%	25%
[94]	2024	MPSO-SA-DQN	75%	0%	25%	25%
[95]	2024	Modified GA	50%	0%	0%	12.5%
[96]	2024	Cooperative Combat Model	87.5%	0%	75%	75%
[97]	2024	MOALO-RSI	50%	0%	75%	25%
[98]	2024	VNS with MILP	25%	0%	0%	12.5%
[99]	2024	IMOGOA	25%	0%	87.5%	25%

presents articles published in 2024,

Table 5. 2023 Articles.

Ref.	Pub. Year	Developed Algorithm	C1	C2	C3	C4
[100]	2023	DCSDM	50%	0%	62.5%	25%
[101]	2023	GMMAA	0%	0%	0%	25%
[65]	2023	IPSO	0%	0%	0%	25%
[102]	2023	IPSO	50%	0%	12.5%	25%
[103]	2023	DDPG	50%	0%	62.5%	25%
[104]	2023	IM-DPSO	25%	0%	0%	25%
[105]	2023	FMASAC	25%	37.5%	12.5%	12.5%
[69]	2023	DQN	0%	25%	12.5%	25%
[106]	2023	EMSSA	0%	50%	62.5%	50%
[107]	2023	PTMA	50%	0%	62.5%	37.5%
[68]	2023	IGA	25%	0%	0%	12.5%
[67]	2023	DA-PSO	50%	0%	12.5%	12.5%
[108]	2023	ILP Auction	25%	0%	0%	12.5%

presents articles published in 2023 and

Table 6. 2022 and Prior Articles.

Ref.	Pub. Year	Developed Algorithm	C1	C2	C3	C4
[109]	2022	PI-Predict	75%	0%	87.5%	87.5%
[64]	2022	QLHH-II	50%	12.5%	25%	12.5%
[110]	2022	Extended CNP	62.5%	87.5%	25%	37.5%
[111]	2022	MILP	75%	0%	0%	12.5%
[112]	2022	EN coupled NSGA-III	50%	0%	75%	12.5%
[113]	2022	IAGA	50%	25%	0%	25%
[114]	2022	MWPSO	25%	0%	0%	25%
[115]	2022	GTO	50%	0%	0%	12.5%
[116]	2021	BCI	25%	37.5%	100%	75%
[117]	2021	D-NSGA3	50%	0%	12.5%	75%
[118]	2021	Advanced GA	0%	37.5%	50%	12.5%
[66]	2021	MOEA/D	0%	0%	37.5%	25%
[119]	2021	FCE	50%	0%	87.5%	25%
[120]	2020	QLHH	25%	0%	37.5%	25%
[121]	2020	Coalition Formation	62.5%	0%	50%	25%
[122]	2018	NSGA-II	62.5%	0%	12.5%	12.5%
[123]	2018	k-PICEA-G	62.5%	0%	0%	12.5%
[124]	2018	DACLD	75%	0%	0%	12.5%
[125]	2018	Improved NSGA-III	50%	0%	37.5%	12.5%
[126]	2013	Genetic Fuzzy Clustering	25%	0%	0%	12.5%
[127]	2007	Genetic Vehicle Routing	0%	25%	0%	12.5%

presents articles published in 2022 and prior.

6. Discussion

6.1. Criterion One Results

Analysis driven by criterion one concluded that the legislative concerns associated with multi-user heterogenous control remain unaddressed by current mission planning systems. Specifically, when algorithms demonstrated distributed online control heterogenous swarms were not considered. In contrast, systems that considered heterogeneous tasks, either required all tasks to be known prior to run time or were implemented using centralized/hybrid control architecture. Another finding of note is that whilst the concept of task heterogeneity is parametrically considered in numerous articles, functionally heterogeneous tasks were rarely discussed.

The root causes behind the highlighted trends were then investigated directly to identify potential enabling technologies that could move toward addressing this area of legislative concern. This investigation started by analyzing the highest scoring articles. Two 2025 articles [80,81] scored 87.5% by implementing hybrid architectures capable of online heterogenous control. Ref. [81] achieved this through the use of an Improved Contract Net Protocol (CNP) algorithm which built upon traditional auction-based approaches to achieve heterogenous control arbitrated by a central UAV node. The work in [80] achieved this through a Task-Driven Clustering (TDC-MOPSO) algorithm which proposes a dynamic cluster head allocation method that also accounted for effective task completion. The 2024 article [96] also implemented a hybrid control structure, this approach leveraged centralized command drones to process mass command and control data to determine optimal task assignments within their respective operational zones. The algorithm demonstrated control heterogeneity by subdividing its task base into reconnaissance and strike tasks. The primary drawback of this approach is that the hybrid structure still relied on a centralized control paradigms as the algorithm effectively treated each of the distributed strike regions as their own unique centralized control problem. Other hybrid approaches such as [110] demonstrated similar restrictions. Reliance on completely centralized control systems also achieved well, such as [111], which was able to facilitate online heterogenous control. Whilst these systems scored well, the algorithms used cannot be transposed into decentralized architectures.

Article [100] again aimed to create a swarm combat strike method, however this article utilized a distributed control system labeled distributed cooperative strike decision method (DCSDM). This paper addressed the core limitation of [96] by implementing an auction-based control topology which allowed for online decentralized control. However, the auctioning method used in this article only considered homogenous tasks. The proposed method allowed all drones to bid on all tasks, given that the bids were only concerned with candidate attack vectors for the given tasks. Ref. [109] demonstrated how bidding structures could be extended to address heterogenous tasks by using a two-stage market auction method. However, in this example the auctioning method required knowledge of all tasks prior to run time. Refs. [94,121] both demonstrated alternative decentralized control structures to the auctioning method used in [100]. In both cases, the functionality of the decentralized control system was extended to satisfy heterogenous control; however, in both cases, the states of all tasks now needed to be known beforehand, indicating that the algorithms could not facilitate online task submission.

6.2. Criterion Two Results

Criterion two’s rubric identified numerous example articles with proposed methods capable of satisfying the legislative requirements if extensions were made. This is reinforced by the fact that many of the analyzed articles omitted obstacle avoidance from their problem models to narrow the developmental focus of the article onto the assignment objectives instead. A consequence of this narrowed scope is that these articles inherently fail to demonstrate sufficient functionality to conform to legislative requirements central to C2. As such could not safely satisfy their functional objectives in real world scenarios. However, a subset of articles did score highly against this criterion using loosely coupled more complex path planning algorithms. It is within these approaches that the inefficiencies associated with loosely coupled subcomponents were highlighted.

The primary delineation between the highest scoring articles and low scoring articles was metrics 2B and 2C. This was highlighted by [82] which scored a criterion high 87.5% by achieving online obstacle detection, distribution and avoidance on a swarm level. Similarly [90] scored 75% by implementing a local detection and avoidance method which was able to maintain avoidance of dynamic obstacles. However, a key downside of this approach is that the obstacle information was only known to the individual detecting unit and was not shared with neighboring swarm agents. A notable limitation of both of these articles was metric 2D, as neither considered online avoidance task submission. This limitation was addressed by [110] which scored 87.5% by demonstrating the ability to facilitate online avoidance task submission, whilst still conforming to 2B and 2C. This was achieved by implementing a tightly coupled mission planning system that integrated obstacle avoidance into the mission planning process using a dynamic survivability estimation, as shown in Figure 7. Notably, the use of a light weight numeric survivability estimate enabled online obstacle avoidance to be integrated within mission planning. Additionally, this approach allowed for accurate fuel assumptions and range calculations (denoted L_m) that accounted for no-fly zone areas (denoted as threats O_m) within the task assignment process (tasks denoted T_m) of each drone (denoted V_m). However, this paper only considered static obstacles. Metric 2A demonstrated that the majority of articles omitted obstacle avoidance systems from their research scopes.

6.3. Criterion Three Results

Results generated by criterion three aligned with previous research made on this aspect of mission planning systems. Key weaknesses highlighted by this criterion’s analysis were a continued reliance on broad communication assumptions to remove risks associated with real world swarm communication scenarios, and a lack of consideration for communication failure. These weaknesses directly compromise a mission planning systems ability to effectively address the command link loss requirements that are currently present within literature. A subset of high-scoring articles presented methods which used minimal communication assumptions and acknowledged the existence of non-perfect communication scenarios, whilst presenting a comprehensive communication model.

The key divide between results was that the highest scoring articles all directly considered the communication network within either their problem description or directly within the modeled optimization problem. Both [84,99] are examples of the latter, scoring 75% by considering communication security directly during optimization. Ref. [80] is an example of the former, whilst not discussed explicitly within the modeled problem, the network characteristics are discussed extensively within the problem description. Whilst none of these three papers scored full marks against criterion three’s rubric, all three papers demonstrated considerable progress toward legislative conformity. Earlier publications such as [87,88,97] also included network optimization parameters directly into the core optimization algorithms within the mission planning system itself. This was reflected in the reduced assumptions these papers made regarding the communication systems. However, whilst these papers did consider communication issues, primarily eavesdroppers, they did not consider communication failure directly. One method used to address communication failure was with an independent auctioneer which arbitrated over task assignment using a confirmation message scheme. This method was introduced within [89]; however, this paper did not validate this method within their simulated experiments. Ref. [96] proposed an alternative approach which modeled the intra-swarm communication network as a connected graph, representing drone communication links as graph edges. However, despite using a graphical method to represent its intra-swarm communication, this paper was one of the few approaches that considered UAV link loss and link repair. The method presented within this paper highlighted similar stability structures to [80], whereby both papers implemented hybrid control architectures and dynamic edge or head allocation methods, as shown in Figure 8. In Figure 8, the control link between the decision drone in position one (D1) and the heterogenous (sensing and decision) drone in position 4 is broken. This is rectified by establishing a new edge between the drones in positions 2 and 4, which maintains network integrity by ensuring all drones have at least one redundant link. By directly considering these dynamic allocation methods within the algorithm design, network stability is greatly improved.

6.4. Criterion Four Results

Analysis driven by criterion four highlighted a lack of consideration toward swarm observability within the review article base. This suggests that future articles may continue to overlook this unique set of legislative requirements. In general, the majority of articles relied on omniscient simulation platforms to analyze algorithm performance. This approach was adopted by a vast majority of the analyzed articles as it is the simplest and arguably most effective method for analyzing the optimization capacity of the proposed solutions. Whilst this was a positive finding regarding the rigor being shown toward performance validation, these validation methods have very little translation into analyzing the swarm’s performance in real world scenarios where an omniscient swarm viewpoint cannot be maintained. As such, reliance on such methods directly hinders these approaches’ ability to conform to the legislative requirements central to criterion four. However, despite an overall reliance on these omniscient simulation platforms, a select few articles did recognize the need for observation methods to allow for real world performance analysis.

As stated, the vast majority of the reviewed papers used simulation platforms to numerically oversee the optimization process of the developed algorithms. Both [80,83] are exceptions to this, however, as they achieved 75% score by supplementing numerical simulation for optimization evaluation with real-world evaluation. Other approaches proposed full simulation environments to evaluate optimization within an environment. Such as the solution presented in [65] which used a Python (version 3.9) based environment to calculate the reward function of each agent, at each time step to quantify swarm performance. Similar approaches were made in [101] which used a C++ direct calculation approach and [94] which also used a Python environment. Another common approach implemented generic pre-existing simulation platforms to evaluate algorithm performance in a less direct manner. These approaches included [86] which implemented the Gazebo ROS simulation tool to validate its algorithm. Ref. [87] took a similar approach implementing an off-the-shelf tool to generate visual simulations and numeric evaluation data. The key limitation of the above approaches is that the simulation platforms inherently know the precise states of all UAVs within the swarm and rely on this information to perform their analysis. Furthermore, none of the above evaluation techniques considered spurious or non-planned phenomena, only evaluating the swarm’s performance under perfect conditions. The highest scoring paper against this criterion, ref. [96] partially addressed these concerns by providing a simulation environment capable of simulating un-planned intra-swarm link failure events which forced the swarm to adapt to non-perfect operating conditions. However, again this simulation platform still relied on an omniscient viewpoint to preform analysis.

7. Open Challenges and Enabling Technologies

7.1. Online Decentralized Heterogenous Multi-User Control

Criterion one’s discussion in Section 6.1 identified two core limitations that are restricting UAV swarms from satisfying the multi-user control considerations presented within legislation. The first, showed that hybrid/centralized task assignment systems remain a popular approach adopted by modern articles, presenting functionality that cannot be transferred to fully distributed control systems. The second, showed that when decision-making authority is maintained onboard each drone, i.e., fully decentralized architectures are used, online heterogeneous control was not achieved. Without a distributed control method, distributed users cannot submit control directly to the swarm, effectively rendering the swarm as a single user system, a reduction which directly contradicts legislative considerations. Furthermore, without online heterogenous control unique users cannot maintain in-the loop control of the swarm post deployment which increases the risks associated with operation.

To address these limitations, future systems would need to facilitate online heterogeneous control whilst maintaining onboard decision-making authority via decentralized command architectures. Potential approaches include modifications to onboard adaptive consensus algorithms [81] such that they can maintain flat command structures to allow for fully decentralized control without requiring a hybrid structure to arbitrate [80]. Considerations would need to be made to extend performance to handle heterogeneous control and to allow for distributed user task assignment. Implementing such methods could present numerous practical upsides including increasing scalability by reducing reliance on centralized infrastructure through distributed control, increasing cost-effectiveness through the added versatility of heterogenous control, and increasing system safety by allowing distributed users to remain in the control loop. All three of these benefits are critical in not only existing UAV markets but also within developing regions where limited infrastructure, budget constraints and requirements for safe, adaptable operations present major barriers to UAV deployment.

7.2. Intergration of Robust Path Planning into Multi-Objective Mission Planning Systems

Criterion two’s discussion in Section 6.2 highlighted the limited application of robust obstacle avoidance methods within multi-objective mission planning systems. Within broader UAV path planning literature these limitations have been addressed directly [59,128]. For example, complex path planning problems such as dynamic obstacle avoidance, obstacle location transmission and group avoidance paradigms have all been investigated in isolation by existing path planning methods [59,128]. As such, the primary open challenge identified by criterion two’s analysis, is not the lack of capable methods but rather their lack of application within multi-objective mission planning systems. The application of these path planning methods needs to be considered during the design of the mission planning system to maintain safe operation and function within the computational limitations of an applied swarm systems. If task assignment and path planning systems continue to be developed in isolation, as is performed when obstacles are omitted from task assignment problem definitions, the final solution exhibits two core limitations. The first is demonstrated by developed solutions which omit obstacles during path planning to reduce problem complexity. Without obstacle avoidance the developed solution cannot conform to legislative considerations. The second is demonstrated by the arduous process required to extend an existing approach to incorporate obstacle avoidance. If a more robust path planning system is to be implemented post task assignment, distributed tasks may become inefficient if paths require excessive adjustment. In contrast if a more robust path planning system is implemented pre-assignment, then the computational requirements of the system could increase exponentially due to the added resources required to generate candidate paths.

To address these limitations, future systems would need to consider inter-component coupling during the design of the multi-objective mission planning system, if a tighter coupling approach is taken, these limitations can be minimized. Currently actionable methods include the implementation of iterative task-path coupling approaches, such as assigning tasks using basic heuristics and then refining them with lightweight path feasibility checks, reducing the need for exhaustive path generation upfront. Alternatively, dynamically updated risk maps (e.g., occupancy grids or costmaps) [128] that can be shared across both task assignment and path planning modules to inform obstacle-aware decisions without complex real-time path-regeneration [128]. These technologies would not only move UAV swarm systems toward legislative requirements, but they would also drastically improve system efficiency, helping to maximize battery life performance during operation.

7.3. Intergation of Robust Networking into Swarm Control Systems

Criterion three’s discussion in Section 6.3, presented similar conclusions to the open challenges identified by criterion two. That is, methods for identifying and adapting to communication failures within networked systems are not novel concepts [1] but their application within UAV swarm systems is. Within static networked systems, many of the assumptions made by the articles discussed in Section 6.3 would likely hold in real world applications [129]. Furthermore, in static systems simply reporting communication failures is often sufficient to ensure the safe shutdown of the networked system. However, due to the relative speed and physical profiles of UAV platforms these assumptions do not hold for UAV swarm systems, and as such cannot satisfy the legislative considerations central to criterion three. Thus, criterion three’s identified open challenge relates to the adaptation of conventional communication system management techniques to satisfy the unique challenges of UAV swarm. This re-iterates the concerns presented with [55,130], highlighting that these challenges remain unaddressed by modern UAV swarm systems.

The analysis performed by this criterion highlighted three areas of interest that will need to be addressed by mission planning systems to satisfy this open challenge. Specifically, future systems will need to develop approaches which clearly define the physical actions required to mitigate the immediate risks of communication failure, develop identification methods which can quickly identify communication faults and develop avenues for network repair. Practical solutions to this could involve the adoption of lightweight, redundant communication links using predefined failover priorities to maintain partial network integrity under failure [130]. Furthermore, by considering the functional implications of periodic heartbeat packets that monitor link health [130] within mission planning systems, when these fail to arrive, UAVs could autonomously initiate predefined fail-safe tasks (e.g., hover, return-to-home, or regroup). Implementing these methods would not only move these systems toward legislative conformity, but could also allow for alternative communication mediums to be used in real time, such as satellite communications, given the added onboard link-lost robustness.

7.4. Swarm Observability Support

The open challenges identified by criterion four’s analysis are directly related to the limitations highlighted in Section 6.4. Specifically, the lack of real-time analysis techniques and lack of support for non-solicited event communication in decentralized systems. As it currently stands, due to these limitations real world swarm observers cannot maintain an informed in the loop station alongside swarm systems, especially if decentralized control architectures are used. This inherently fails to address the legislative considerations central to criterion four. To achieve legislative conformity these limitations must be addressed directly. This would require the implementation of accurate mission status reporting methods within mission planning systems coupled with the development of simulation platforms which use omniscient viewpoints to validate not only the algorithm performance but also the accuracy of the swarm’s returned information.

Future systems aiming to address these requirements could implement event-driven data pipelines that could allow individual UAVs to push unsolicited mission-critical updates (e.g., deviations, task failures) to external observers in real time. Additionally, local aggregators could be implemented to summarize status information from nearby UAVs and periodically relay that information to operators. An immediate extension that can be made to the omniscient simulation platforms currently implemented is the introduction of observer viewpoints. These viewpoints would allow relayed mission status information to be validated directly. By coupling these approaches authors will be able to not only provide critical operation information on the swarm but also validate the accuracy of said information, which will directly address the core tenets presented in criterion four’s legislative considerations.

8. Conclusions

Using a systematic review, this paper identified key gaps between the functional considerations proposed within legislation and the current functionality exhibited by swarm systems. The systematic review process applied four unique criteria to identify subsets of requirements that remained unaddressed by the current technical capabilities of multi-objective mission planning systems. Each criterion investigated a current legislative consideration, namely UAS Instruction from multiple external observers, dynamic no-fly-zone geofencing, command and control link loss, and UAV swarm observability. These results were then used to identify the technical factors limiting UAV swarm systems from addressing these requirements and to propose open challenges aimed at addressing these limitations. Therefore, this review successfully achieved its core objectives by (1) applying a novel systematic framework targeted at a unique research perspective, (2) identifying key discrepancies between UAV swarm functional capacity and legislative concern that are limiting UAV swarm application, and (3) proposing targeted research directions aimed at bridging this divide. This review process had two scope constraints of note: (1) this review was limited to multi-objective mission planning systems, and (2) it focused on a selected subset of active legislative items proposed by JARUS. Future work could supplement these constraints by applying this reviews framework to other UAV swarm subsystems, or to a broader set of legislative requirements.