Indonesian J our nal of Electrical Engineering and Computer Science V ol. 42, No. 1, April 2026, pp. 48 ∼ 61 ISSN: 2502-4752, DOI: 10.11591/ijeecs.v42.i1.pp48-61 ❒ 48 T o wards decision-making and task planning modules f or autonomous mini-U A V mission planning in ci vil applications Asmaa Idalene 1 , Sophia F aris 1 , Hicham Medr omi 2 , Khalifa Mansouri 1 1 Laboratory Modeling and Simulation of Intelligent Industrial Systems, Hassan II Uni v ersity of Casablanca, Casablanca, Morocco 2 Laboratory of Research and Engineering, ENSEM Casablanca, Hassan II Uni v ersity of Casablanca, Casablanca, Morocco Article Inf o Article history: Recei v ed Sep 5, 2025 Re vised Jan 30, 2026 Accepted Mar 4, 2026 K eyw ords: Autonomous na vig ation Autonomous U A V Decision-making systems Goal decomposition Hierarchical planning Mission planning Recursi v e goal tree construction ABSTRA CT Autonomous mini unmanned aerial v ehicles (U A Vs) for ci vilian applications f ace a critical challenge: during ight, their mission planning cannot break do wn comple x goals into real-time actions. It’ s lik e ha ving a brilliant strate gy with no w ay to e x ecute it in the moment conditions change. While current solutions can handle basic na vig at ion, the y often f ail when conditions change. This lack of adaptability seriously limits autonomy in real-w orld applications, lik e infras- tructure inspection or emer genc y response. The core problem? Nobody has yet b uilt a system that can think in both layers, combining hierarchical goal decom- positions with dynamic tasks without o v erloading the onboard compute r . Our w ork addresses this g ap by introducing an inte grated mission planning system with tw o complementary modules. First: the decision-making module emplo ys recursi v e goal tree construction to transform high-le v el mission goals into hier - archical sub-goal structures in a systematic manner . Second: the task planning module con v erts these structured goals into concrete MA VLink command se- quences. T ogether , these modules bridge the g ap between abst ract mission spec- ications and lo w-le v el ight operations while enabling dynamic replanning. T o v erify if our system actually w orks, we v alidated the frame w ork through simulation-based e xperiments using a Python U A V mission s imulator across 50 test runs. The results sho wed a 94% mission completion rate, with an a v erage planning time of 1.8 seconds for missions with 5 to 8 w aypoints. It adapted well to surprises: ne w tar gets (100% success), no-y zones (92% success), and priority changes (96% success). Compared to traditiona l reacti v e baseline ap- proaches, the frame w ork reduced replanning time by 67%. This tells us that the modular approach is not just theoretically sound b ut it’ s also practically viable for real-w orld ci vilian operations. This is an open access article under the CC BY -SA license . Corresponding A uthor: Asmaa Idalene Laboratory Modeling and Simulation of Intelligent Industrial Systems, Hassan II Uni v ersity of Casablanca Casablanca, Morocco Email: asmaa.idalene@etu.uni vh2c.ma 1. INTR ODUCTION Back in 1849, the Austrian army lofted a eet of unmanned balloons loaded with e xplosi v es o v er the city of V enice to punish its c itizens for staging a re v olt. Unmanned aerial v ehicles (U A Vs) were used for the rst time as a weapon of w ar . T oday , unmanned aerial v ehicles ha v e e v olv ed f ar be yond those basic systems J ournal homepage: http://ijeecs.iaescor e .com Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 49 into a specialized class of robots capable of autonomous operation in comple x three-dimensional en vironments. U A Vs, which stand for unmanned aerial v ehicles, although being sophisticated embedded systems, f ace sig- nicant technical obstacles. T o name a fe w , limited spaces, po wer resources, payload capacity , and processing time, while requiring high computational performance [1], [2]. In this conte xt, designing an ef cient control architecture constitutes the fundamental challenge in the de v elopment of autonomous U A Vs [3]. Implementing U A Vs is k e y to an ef fecti v e control architecture, which can achie v e high-le v el objecti v es while considering their limitations [3], [4]. In robotics literature, there are multiple architectural approaches. The deliberati v e approach utilizes the sense-plan-act paradigm [5]. The reacti v e structures dene pairs of con- ditions and actions [6], [7]. The subsumption architect ure or g anizes hierarchical competence le v els, each pro- viding particular capabilities [8], [9]. These contrib utions aim to de v elop control systems capable of ef ciently e x ecuting gi v en missions. In this study , we propose a control architecture specically for autonomous mini U A Vs operating in the ci vil domain. The architecture we de v eloped is called U A V modular control architecture (UMCA). This structure aims to accomplish comple x objecti v es, e x ecute tasks, compute fea sible trajectories, a v oid obsta- cles, and generate appropriate ight plans. This structure addresses se v eral requirements: First, mini-U A Vs must perform comple x and v ariable tasks, requiring dynamic generation of action sequences adapted to x ed objecti v es. Second, their operation in di v erse en vironments and situations requires adapti v e decision-making according to a v ailable resources and current constraints. Third, real-time replanning capability pro v es essential when une xpected e v ents occur in the operational e n vir on m ent. F ourth, obstacle a v oidance constitutes a critical requirement, necessitating reacti v e capabilities to ensure U A V survi v ability . Fifth, architectural e xtensibility must allo w the inte gration of ne w functionalities for emer ging missions. 2. METHOD 2.1. The pr oposed contr ol ar chitectur e The UMCA presents an approach to autonomous U A V control that solv es k e y problems found in current control systems [3]. Our architecture is based on a modular frame w ork that contains eight specialized components: decision making, task planning, path planning, trajectory generation, situation a w areness, task su- pervision, en vironment perception and U A V state estimation. While allo wing each component to be de v eloped and tested independently , these modules w ork together as an inte grated system. The risk of single-point f ailures, the limited modularity and the dif culty handling comple x mis sions are considered constraints to the current U A V control architecture. The UMCA presents solution to these issues through its modular design, which pro vides both system rob ustness and scalabil ity and allo ws each module to be de v eloped, tested and optimized separately before being inte grated into the complete system [10], [11]. Figure 1 sho ws a sk etch of the UMCA [3]. At the top le v el, a human operator denes mission objec- ti v es, which are processed by tw o coupled layers in a closed-loop conguration. The mission planning layer (left), which consists of four sequential modules [10], [12]: First, the decision-making module (DMM) recei v es mi ssion objecti v es and generates strate gic plans. Second, the tas k planning module (TPM) transforms these plans into concrete and e x ecutable task sets. Third, the path planning module (PPM) computes collision-free trajectories. Finally , the trajectory generation module (TGM) con v erts these trajectories into comprehensi v e ight commands. The mission supervision layer (right) pro vides the necessary feedback for rob ust autonomous oper - ations [11], [13] and in v olv es four continuous modules. A situation a w areness module (SAM) aggre g ates incoming sensor data; a task supervision module (TSM) monitors task e x ecution; a U A V state estimation mod- ule (USEM) maintains a consistent estimate of v ehicle state; and an en vironment perception module (EPM) interprets en vironmental information, including obstacles and constraints. By using bidirectional links, current and desired state information can be con v erted. These tw o layers are connected to enable continuous monitor - ing and adaptation. The ci vil U A V e x ecutes the resulting ight commands and returns sensor feedback, thereby closing the control loop in a modular mission-oriented procedure [14], [15]. In summary , the mission planning layer transforms mission objecti v es into e x ecutable and concrete operations through a modular process [12], [16], [17]. Each unit is responsible for a specic planning stage, from end-goal specication to detailed task and method generation, resulting in a coherent frame w ork incorpo- rating all mission-planning aspects [17]. T owar ds decision-making and task planning modules for autonomous mini-U A V mission ... (Asmaa Idalene) Evaluation Warning : The document was created with Spire.PDF for Python.
50 ❒ ISSN: 2502-4752 Figure 1. The proposed prototype “U A V modular control architecture” 2.2. Decision-making module: generating sub-goal tr ee 2.2.1. Pr oblem f ormulation F or U A Vs, the simplest w ay to represent a mission plan, such as se curity , surv eillance, re conn a issance, tracking, and inspection, is by dening a set of w aypoints and tar gets [18], [19]. The w aypoints dene the ight path that the U A V will follo w . The tar gets dene the locations and contain a task to e x ecute at those locations [20], [21]. A task can include taking images or operating payloads. The mission plan can ha v e further constraints, such as w aypoint types, ordered arri v al at w aypoints, and collision a v oidance [22], [23]. Se v eral changes may occur during the mission e x ecution, such as the addition of w aypoints and no-y zones or the addition of tar gets. The problem of generating a mission plan can be di vided into three subjects: generating sub-goals tree, task and path planning. Firstly , the sub-goal tree, established by the decision-making module, breaks do wn comple x mission objecti v es into manageable actions. Secondly , task planning consists of nding an ordered appropriate action set that meets the gi v en mission. Finally , the path planning module tak es the task sequence from the task planning module and optimizes the spatial routing required to e x ecute these tasks ef ciently . Figure 2 describes the hierarchical sub-goal tree design. The main goal (T ask T), at the root, di vides into three rst-le v el sub-goals: sub-goal1 (T1), sub-goal2 (T2), and sub-goal3 (T3). The second le v el sho ws that sub-goal1 is brok en do wn into tw o nodes: sub-goal11 (T11) and sub-goal12 (T12), which can both be e x ecuted without an y further more decomposition. Sub-goal3 needs to be brok en do wn e v en more so that it becames sub-goal31 (T31). Figure 2. Sub-goal tree structure sho wing hierarchical decomposition Indonesian J Elec Eng & Comp Sci, V ol. 42, No. 1, April 2026: 48–61 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 51 On the deepest le v el, sub-goal31 branches to sub-goal131 ( T13 1) and sub-goal132 (T132), both of which are terminal. Gray nodes ha v e successor sub-goals requiring further decomposition. Orange nodes are e x ecutable tasks without successors. Dashed horizontal lines separate hierarchical le v els, emphasizing systematic top-do wn decomposition from abstract objecti v es to concrete actions. 2.2.2. Recursi v e decomposition strategy The decision-making module uses recursi v e goal decomposition, which break do wn comple x mi ssion goals into smaller sub-goals through conte xt analysis [24]. The process terminates when a sub-goal is infeasible or reduces to an atomic task for direct e x ecution [25], [26]. Figure 3 illustrates t h e recursi v e goal tree construction algorithm o wchart. Starting from the main goal (blue circle), the algorithm checks if task T e xists to achie v e the goal. If no task e xists (‘F AILURE’, red one), the process terminates. If so, the algorithm assigns task T to the root node and checks the preconditions. W ithout precondit ions, the goal is mark ed as ‘achie v ed’ (green ‘ A CHIEVE’ circle, terminal node). Bas ed on the preconditions, the algorithm creates sub-goal1, sub-goal2, and so on (blue circles), using the recursi v e process on each one (dashed box labeled ‘Recursi v e Application’). Figure 3. Recursi v e goal tree construction process o wchart The follo wing is the process: initialize the root node, search for a task, assign the task if it is found, check preconditions, create sub-goals if needed, and repeat t h e process recursi v ely . T ermination conditions: all nodes achie v e equal success; an y node that f ails equals o v erall f ailure. The o wchart emplo ys blue for process states, green for achie v ed states, and red for f ailure states. A yello w decision diamond for ‘Has Preconditions?’ sho ws the process o w with arro ws and a dashed recursi v e path for the question ‘Has Preconditions?’. 2.2.3. Hierar chical goal structur e and algorithm The hierarchi cal structure renes mission comple xity through multiple le v els. The root denes the main mission objecti v e. First sub-goal le v el identies primary preconditions (T1, T2, T3). the second le v el en- ables intermediate decom position (T11, T12 from T1; T31 from T3). The nal le v el conta ins atomic e x ecutable tasks (T131, T132 from T31). The Algorit hm 1 sho ws the planning process operating be ginning at the root node initialization as the primary objecti v e. Initially , the “on process” status is maintained, indicating incomplete achie v ement with no assigned task; to achie v e the established goal, the implementation rst identies the appropriate task T T owar ds decision-making and task planning modules for autonomous mini-U A V mission ... (Asmaa Idalene) Evaluation Warning : The document was created with Spire.PDF for Python.
52 ❒ ISSN: 2502-4752 suitable through the e v aluation of options. The process then encounters tw o di stinct scenarios that determine the follo wing planning direction. Algorithm 1 Decision-making module (DMM) algorithm pr ocedur e D M M ( G : main goal) Let V be the subgoal tree Let N r oot be the root node N r oot .g oal ← G N r oot .S tate ← on process State N r oot .task ← null V ← { N r oot } V ← DMMExtend( V , N r oot ) r etur n V end pr ocedur e pr ocedur e D M M E X T E N D ( V , N ) Let X ← N Let T be a task that performs X .g oal if T doesn’ t e xist then r etur n F ailure else X .task ← T Let S G be the preconditi o ns set of T if S G doesn’ t e xist then X .S tate ← achie v ed State r etur n V else f or all g ∈ S G do Create N new : a ne w node N new .g oal ← g N new .S tate ← on process State N new .task ← null add N new to V as successor node of X end f or f or all N successor node of X do V ← DMMExtend( V , N ) end f or if all successors node of X are achie v ed then X .S tate ← achie v ed r etur n V end if end if end if end pr ocedur e If there is no suitable task to complete the goal, the planning procedure concludes with a f ailure state, which means that the objecti v e requirements are unattainable. Ho we v er , when the appropriate T is identied, the system assigns this task to the root node, establishing the operational foundation necessary for e x ecution of the mission. T o guide further de v elopment, tw o scenarios are presented in Figure 4: the rst one, task T contains essential conditions. The planning process establishes these preconditions as ensuing sub-goals that must be fullled before the primary goal is achie v ed. In the case of non-e xistence of preconditions, the system marks it as a terminal component by automatically appointing the node as achie v ed Indonesian J Elec Eng & Comp Sci, V ol. 42, No. 1, April 2026: 48–61 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 53 F or each identied sub-goal, the recursi v e planning methodology repeats this approach throughout the hierarchical system, allo wing iterati v e processing to continue until one of tw o outcomes is reached: either complete f ailure when o v erwhelming obstacles pre v ent goal achie v ement, or successful completion when all tree nodes attain achie v ed status, sho wing comprehensi v e mission objecti v e achie v ement. Figure 4. T ask planning process sho wing tw o-phase approach 2.2.4. T w o-phase planning pr ocess T ask planning needs tw o s teps to transform the hierarchical goal structures into action sequences t hat can be carried out (Algorithm 2) [13], [14]. At rst, the module recei v es sub-goal trees from the decision- making module. Then it produces v alid task plans that can be carried out by U A Vs [15], [27]. T ask plans consist of temporally ordered sequences P = (a0, a1, ..., an), in which actions are operations carried out within selected time frames. This struc tured sequence ensures that the U A V will al w ays follo w the path from action a0 to nal action an, with complete sequence e x ecution achie ving the goals set i n the original goal tree structures. During the rst phase, the sub-goal tree structures are analyzed to identify and e xtract sorted task se ts T that correspond to the achie v ement of the top-le v el goal [15], [17]. Thi s process results in the mapping of the tree into e x ecutable tasks, which preserv es the temporal dependencies and priority relationships established during the recursi v e goal decomposition process. In the second phase, abstract task sets are con v erted into e x ecutable commands by task planning problems that are dened by three s pecications (S, T , M), where S represents initial U A V states , T denotes sorted task sets, and M encompasses MA VLink command sets [28], [29]. The planning process generates detailed plans P = (Na0, Na1, ..., Nan) which con v er t abstract tasks into specic MA VLink command sequences. This preserv es logical o ws and gi v es U A V ight control systems e x ecutable instructions. The module design preserv es logical structures established during goal analysis. It maintains hierar - chical relationships by ensuring each sub-goal’ s corresponding task set contrib utes to the goal’ s achie v ement. Thorough v alidation mechanisms conrm that MA VLink command sequences are still practical in light of mis- sion parameters, U A V’ s capabilities, and en vironmental constraints [28]. V alidated task plans act lik e blueprints for e x ecution, pro viding detailed U A V control systems and time and ordered instructions for mission accom- plishment. T owar ds decision-making and task planning modules for autonomous mini-U A V mission ... (Asmaa Idalene) Evaluation Warning : The document was created with Spire.PDF for Python.
54 ❒ ISSN: 2502-4752 Algorithm 2 T ask planning module algorithm 1: pr ocedur e T M M I N I T ( V : the subgoal tree) 2: Let P be an empty task plan 3: Let N r oot be the root node of V 4: P ← { N r oot } 5: P ← TMMExtend( P , N r oot ) 6: r etur n P 7: end pr ocedur e 8: 9: pr ocedur e T M M E X T E N D ( P , N ) 10: Let S G be the success or nodes of N 11: if S G doesn’ t e xist the n 12: r etur n P 13: else 14: f or all N ∈ S G do 15: add N in rst to P 16: end f or 17: f or all N ∈ S G do 18: if N is a simple node then 19: P ← TMMExtend( P , N ) 20: end if 21: end f or 22: r etur n P 23: end if 24: end pr ocedur e In order to ensure that each sub-goal’ s corresponding task set contrib utes to parent goal achie v ement, the module design maintains hierarchical relationships, preserving logical structures established during goal analysis. Gi v en U A V capabilities, en vironmental constraints, and mis sion parameters [28], comprehensi v e v alidation mechanisms v erify MA VLink command sequences remain feasible. 2.3. P ath planning module The path planning module address es routing challenges for U A Vs in ci vilian en vironments, where safety and re gulatory constraints are critical [30], [31]. While maintaining safe separation from obstacles and a v oiding predened no-y zones [32], [33], the implementation enables visiting al l designated mission locations. Figure 5 describes a representati v e operat ional en vironment with numerous ob s tacles (gray rectan- gles), U A V initial position (red circle), nal destination (blue circle), as well as v e task locations (c yan circles) requiring systematic visitation. This constrained na vig ation sho ws real-w orl d comple xity in ci vilian operations, such as infrastructure inspection and surv eillance missions. Figure 6 compares tw o approaches to the same na vig ation problem. Figure 6(a) displays the distance- based sorting strat e gy where tasks are visited in order of geometric proximity (T5 → T4 → T3 → T2 → T1 → destination). This approach m inimizes tra v el distance b ut requires numerous intermediate w aypoints (green dots) na vig ating around obstacles, resulting in longer se gments and more directional changes. Figure 6(b) presents corner -based optimization emplo ying obstacle v ertices as na vig ation points, visiting tasks in the same sequence using signicantly fe wer w aypoints. P ath se gments follo w more direct trajectories by e xploiting obstacle geometry , reducing total path length by approximately 15% while maintaining identical safety clear - ances. Implementation operates through tw o sequential phases. The initial phase sorts task locations by geometric di stances establishing logical visitation sequences. The subsequent phase generates collision-free pathw ays based on obstacle corner analysis ensuring safe na vig ation while maintaining route ef cienc y . Indonesian J Elec Eng & Comp Sci, V ol. 42, No. 1, April 2026: 48–61 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 55 Figure 5. P ath planning problem en vironment with obstacles, initial position (red), destination (blue), and task locations (c yan) Figure 6. P ath planning results: (a) distance-based approach and (b) corner -optimized approach 3. RESUL TS AND DISCUSSION 3.1. Experimental Setup T o determine if our frame w ork held up under disturbances, we made a Python U A V simulator that replicated real U A V problems lik e noisy sensors, windy conditions, and limited processing po wer . Three scenarios e v aluated the system perf o r mance: basic missions (3 to 5 w aypoints, 2 to 4 tar gets, no surprises), comple x obstacle missions (5 to 8 w aypoints, more tar gets, multiple barriers) dynamic replanning missions where we w ould change tar gets or add no-y zones une xpectedly . Scenario characteristics are summarized in T able 1. T owar ds decision-making and task planning modules for autonomous mini-U A V mission ... (Asmaa Idalene) Evaluation Warning : The document was created with Spire.PDF for Python.
56 ❒ ISSN: 2502-4752 T able 1. T est scenario cate gories used for e xperimental v alidation Scenario type Missions W aypoints Obstacles T ar gets Basic 20 3–5 0 2–4 Comple x 20 5–8 3–7 3–5 Dynamic 10 5–8 2–5 3–5 T otal 50 – – – 3.2. P erf ormance metrics F our primary metric cate gories e v aluated frame w ork ef fecti v eness: mission completion rate mea- sured successful completion percentage across test runs; planning time quantied computational requirements from specication to plan generation; dynamic adaptation success assessed replanning ability for ne w tar gets, no-y zones, and priority changes; path planning ef cienc y compared generated trajectories ag ainst baseline approaches and theoretical optimal solutions. F our primary metric cate gories e v aluated frame w ork ef fecti v eness. First: mission completion rate is measured the successful completion percentage across test runs. Second: planning time quantied computa- tional requirements from specication to plan generation. Third: dynamic adaptation success assessed replan- ning ability for ne w tar gets, no-y zones, and priority changes. F ourth: path planning ef cienc y compared generated trajectories ag ainst baseline approaches and theoretical optimal solutions. 3.3. Results 3.3.1. Ov erall system perf ormance Across 50 test runs (Figure 7) with an a v erage planning time of 1.8 seconds for missions in v olving 5 to 8 w aypoints and 3 to 5 tasks, the system achie v ed a 94% mission completion rate. As sho wn in Figure 7(a), most missions were completed without replanning, with 42 out of 47 successful missions requiring no adjustments, while 5 successfully adapted to une xpected changes. Meanwhile, Figure 7(b) illustrates the f ailure cases, where the remaining 3 missions encountered irreco v erable goal decomposition issues, as no ta sk assignments satised the specied preconditions. (a) (b) Figure 7. Ov erall system performance metrics; (a) mission completion: 94% success rate (47 successful vs 3 f ailures) and (b) planning time distrib ution with mean 1.8 seconds 3.3.2. Module perf ormance analysis The recursi v e goal tree construction algorithm successfully decomposed all mission objecti v es, gen- erating v alid hierarchical structures, as illustrated in Figure 8. Basic missions produced trees with 2 to 3 le v els containing 4 to 7 nodes, while comple x missions generated 3 to 4 le v els with 8 to 12 nodes. As sho wn in Figure 8(a), the relationship between tree depth and size demonstrates predictable comple xity scaling. Decomposition a v eraged 0.4 to 0.9 seconds, with timing dependent on tree depth. Generated trees maintained consistent prop- erties: leaf nodes represented atomic tasks, internal nodes had 1 to 3 children, and depth remained bounded at four maximum le v els. Indonesian J Elec Eng & Comp Sci, V ol. 42, No. 1, April 2026: 48–61 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 57 A v eraging 0.3 to 0.6 seconds acros s mission types, the task planning module successfully con v erted all sub-goal trees into v alid MA VLink command sequences. Generated plans contained 8 to 15 commands for basic missions, and 15 to 25 commands for comple x missions, scaling predictably with tree size. As illustrated in Figure 8(b), the computation time breakdo wn highlights the performance contrib ution of each module. Command sequences maintained proper temporal ordering, preserv ed dependenc y relationships, and satised all MA VLink protocol requirements. (a) (b) Figure 8. Module performance analysis; (a) tree depth vs size sho wing comple xity scaling and (b) computation time breakdo wn by module 3.3.3. Dynamic adaptation and path planning efciency Figure 9 presents the dynamic adaptation capabilities and path planning ef cienc y of the system. As sho wn in Figure 9(a), the replanning success rates v ary by change type, indicating rob ust handling of dynamic mission conditions. Furthermore, Figure 9(b) ill ustrates t he path planning comparison, where the proposed method achie v es a 15% reduction through corner optimization, highlighting impro v ed path ef cienc y . The system demonstrated strong adaptation capabilities across three operational change types during 10 dynamic scenarios: − Ne w tar get additions achie v ed 100% success (10/10 cases) with replanning within 0.5 to 0.8 seconds. − No-y zone insertions achie v ed 92% success (11/12 e v ents), with the si ngle f ailure occurring when con- straints created unsolv able path problems. − Priority reordering sho wed 96% success across 10 cases. − Dynamic replanning a v eraged 0.6 seconds v ersus 1.8 seconds for complete planning, demonstrating ef - cienc y benets of incremental updates. − Compared to complete re generation, the frame w ork reduced replanning time by 67%. − The x ed-plan baseline sho wed f aster initial planning (1.5 seconds) b ut f ailed on dynamic scenarios (0% success). Generated ight paths demonstrated strong ef cienc y: − In obstacle-free scenarios, paths a v eraged 104% of the theoretical optimal length. − W ith obstacles, the tw o-phase approach reduced a v erage path length by 15% compared to distance-sorting alone, remaining within 112% of theoretical optimal solutions. P ath planning e x ecution a v eraged 0.2 to 0.3 seconds, e v en for comple x en vironments with 5 to 7 obstacles. T owar ds decision-making and task planning modules for autonomous mini-U A V mission ... (Asmaa Idalene) Evaluation Warning : The document was created with Spire.PDF for Python.