Every day, billions of smartphones, wearables, and connected vehicles quietly harvest data about the world around them—traffic flows, air quality, noise maps, environmental conditions—all contributing to what researchers call Mobile Crowdsensing, one of the most consequential paradigms to emerge from the Internet of Things. The promise is enormous: by leveraging the sensors already carried in people’s pockets, cities and industries can gather large-scale data without deploying expensive dedicated infrastructure. But the promise collides with a stubborn technical reality. The people carrying those devices keep moving, the wireless environment keeps shifting, and the computing resources scattered across edge devices, fog nodes, and remote cloud datacenters keep fluctuating. Deciding which sensing task should run, on whose device, and at which tier of the computing hierarchy—when everything is in motion—remains one of the hardest open problems in real-world crowdsensing deployments. A new study published in Cluster Computing by Seyed Ali Hosseini, Omid SojodiShijani, and Vahid Khajehvand of Qazvin Islamic Azad University in Iran now offers a strikingly effective answer: a bio-inspired, multi-objective optimization framework called MOFARO, which the authors report cuts execution time by 12 to 25 percent, energy consumption by 9 to 20 percent, costs by 11 to 27 percent, and improves Quality of Service by 13 to 25 percent compared with established baseline methods.
The name MOFARO stands for Multi-Objective Fractal-based Artificial Rabbit Optimization, and the framework’s lineage traces back to a relatively recent addition to the metaheuristic zoo. Artificial Rabbit Optimization, introduced in 2022 by Wang and colleagues in Engineering Applications of Artificial Intelligence, mimics the survival strategies of rabbits in nature—specifically their tendency to forage in unpredictable, zigzag patterns that confuse predators, and their behavior of gathering around the richest food sources. Translated into algorithmic language, candidate solutions “hop” through the search space with exploratory moves that resemble fractal-like stochastic wandering, then converge on promising regions identified by the swarm’s collective experience. The original algorithm was designed for single-objective engineering problems; what the Qazvin team has done is rebuild it from the ground up for the messy, competing demands of dynamic task allocation, where no single metric can capture what makes a scheduling decision “good.”
The problem MOFARO tackles is formally brutal. In an edge-fog-cloud crowdsensing system, sensing tasks arrive continuously and are modeled as Directed Acyclic Graphs, or DAGs, in which each node represents a subtask and each edge captures a precedence constraint—subtask B cannot begin until subtask A finishes and its data has been transferred. Assigning these interdependent DAG tasks across a three-tier architecture means contending with wildly heterogeneous resources: user devices at the edge have limited battery and intermittent connectivity, fog nodes such as base stations and gateways offer moderate compute close to the data source, and cloud datacenters provide abundant processing power but impose network latency and monetary cost. Compounding the difficulty, participants in a crowdsensing campaign are mobile users whose future locations are unknown. MOFARO addresses this by incorporating Markov chain mobility predictions, which estimate the probability that a given participant will be within range of a given service area at a given time, allowing the scheduler to allocate tasks proactively rather than reactively to devices that may soon vanish from the network. Experimental constraints—battery state, processing capacity, transmission bandwidth—are folded directly into the optimization model rather than treated as afterthoughts.
What truly distinguishes MOFARO is its treatment of objectives. Classical schedulers collapse everything into a single weighted cost function, a convenience that hides trade-offs: shaving milliseconds off execution time might burn through batteries or inflate cloud rental charges. MOFARO instead treats execution time, energy consumption, monetary cost, and Quality of Service violations as four simultaneous objectives to be minimized, generating a Pareto front—a set of solutions in which no objective can be improved without degrading another. This is the same philosophy that underpins celebrated multi-objective algorithms such as the Non-dominated Sorting Genetic Algorithm II (NSGA-II), developed by Deb and colleagues in 2002, which the authors use as one of their comparison baselines. But where genetic algorithms rely on crossover and mutation operators that can struggle in high-dimensional scheduling spaces, MOFARO’s rabbit-inspired search exploits fractal-based exploration to maintain diversity across the Pareto front, coupled with mechanisms that preserve elitist, non-dominated solutions across iterations.
The experimental evaluation is thorough by the standards of scheduling research. The team simulated established scientific and industrial workflows—the kind of DAG-structured applications widely used as benchmarks since Topcuoglu, Hariri, and Wu’s foundational 2002 work on heterogeneous computing schedulers—deployed across simulated edge-fog-cloud topologies with dynamic user mobility. Performance was assessed using the hypervolume indicator, a standard measure of Pareto front quality that quantifies the volume of objective space dominated by the discovered solution set; MOFARO achieved a hypervolume of 0.85, indicating that its trade-off surface covers a substantial and well-distributed region of the achievable objective space. Beyond hypervolume, the framework outperformed state-of-the-art methods by 1.8 to 3.9 percent in task success rate—a figure that may sound modest but, in a domain where a single failed sensing task can leave a permanent hole in a city’s environmental dataset, translates into materially more complete and trustworthy data collection campaigns.
The authors also subjected their results to statistical rigor, employing the Wilcoxon signed-rank test—the non-parametric hypothesis test, described in Rey and Neuhäuser’s 2011 encyclopedia entry, that is the field’s workhorse for comparing paired algorithmic performance distributions—to confirm that the observed improvements over competing methods are statistically significant rather than artifacts of random variation across simulation runs. This matters in an area of research where sweeping claims sometimes rest on a handful of favorable runs. The baselines against which MOFARO was measured represent the current spectrum of approaches: deep reinforcement learning schedulers such as those explored by Li, Ota, and Dong for fog-based crowdsensing and later work by Wang and colleagues on optimizing system load and response time; hybrid GA-PSO schemes for multi-objective fog scheduling; improved fuzzy logic approaches for dynamic task allocation; and energy-focused refinements of artificial rabbit optimization itself, such as the improved variant proposed by Ghafari and Mansouri in this same journal in 2024.
Why does a bio-inspired metaheuristic outperform learning-based alternatives in this domain? The authors’ results add to a growing body of evidence that, for combinatorial scheduling problems with tight constraints and multiple conflicting objectives, well-designed metaheuristics offer a compelling balance of solution quality, interpretability, and computational cost. Deep reinforcement learning schedulers must be trained, are sensitive to changes in the environment distribution, and provide little transparency into why a particular allocation was chosen—problematic when an operator needs to audit why a safety-relevant sensing task was delayed. MOFARO, by contrast, adapts its search directly to each problem instance, and its Pareto front output gives operators genuine choice: a city agency running an air-quality campaign can select a solution weighted toward battery conservation, while a commercial traffic-monitoring deployment might prioritize latency, all from a single optimization run.
The practical implications ripple outward. Mobile Crowdsensing underpins applications that have quietly become part of urban life—real-time traffic estimation, parking availability maps, environmental monitoring networks, and disaster-response systems that harvest situational awareness from survivors’ phones. In each case, the economics hinge on scheduling efficiency. Every percentage point of energy savings extends participant battery life and, by extension, the longevity and willingness of the crowd itself; every reduction in QoS violations improves the spatial and temporal completeness of the resulting data products. The 2025 literature the authors survey—covering incentive mechanisms for data quality, lottery-based participant rewards, and quality-aware allocation schemes—makes clear that task allocation is the load-bearing wall of the entire crowdsensing edifice: no incentive scheme can compensate for a scheduler that assigns tasks to devices that are about to disconnect or run out of power.
The research also slots into a broader, fast-moving conversation about computation offloading across the edge-fog-cloud continuum. Surveys published in 2025, including Rahmani and colleagues’ comprehensive review of metaheuristic offloading strategies and Patsias and colleagues’ systematic review of edge task allocation, document an explosion of techniques but also a persistent gap: most methods optimize one or two objectives and assume static resource availability. MOFARO’s contribution—modeling tasks as DAGs, predicting user mobility with Markov chains, and jointly optimizing four objectives in a single adaptive framework—represents precisely the kind of integration that survey authors have been calling for. It is also notable that the authors have released their code and datasets in a public GitHub repository, a practice that remains inconsistent in the scheduling literature and that will allow independent teams to verify the reported gains and build on the framework.
As sensor-laden devices multiply and smart-city infrastructures grow more dependent on crowdsourced data, the scheduling layer beneath these systems will only grow in importance. MOFARO demonstrates that the answer to one of the Internet of Things’ most stubborn dynamic-allocation problems may come not from ever-larger neural networks but from a carefully engineered swarm of virtual rabbits, hopping fractally through the solution space toward better, faster, cheaper, and greener sensing. Whether the framework’s 12-to-27-percent improvements hold up under the noise of production deployments—at true city scale, with adversarial mobility patterns and adversarial participants—will be the next test. But for now, the Qazvin team’s results mark a meaningful step toward crowdsensing systems that can keep pace with the restless, unpredictable human world they are built to measure.
Subject of Research: Multi-objective optimization of dynamic task allocation in edge-fog-cloud Mobile Crowdsensing systems using a fractal-based Artificial Rabbit Optimization algorithm
Subject of Research: Technology and Engineering
Article Title: MOFARO: A multi-objective optimization framework for dynamic task allocation in edge-fog-cloud mobile crowdsensing systems
Article References: Hosseini, S. A., SojodiShijani, O., & Khajehvand, V. (2026). MOFARO: A multi-objective optimization framework for dynamic task allocation in edge-fog-cloud mobile crowdsensing systems. Cluster Computing, 29(12), Article 720. https://doi.org/10.1007/s10586-026-06473-3
Image Credits: AI Generated
DOI: 10.1007/s10586-026-06473-3
Keywords: Mobile Crowdsensing, Task Allocation, Multi-Objective Optimization, Fog Computing, Edge Computing, Task Offloading, Resource Management, Internet of Things, Artificial Rabbit Optimization, Pareto Front
Cite Scienmag News
APA MLA Chicago
Denise Maddox. (September 3, 2026). New framework optimizes dynamic task allocation across edge-fog-cloud crowdsensing systems. Scienmag. https://scienmag.com/new-framework-optimizes-dynamic-task-allocation-across-edge-fog-cloud-crowdsensing-systems/
Denise Maddox. “New framework optimizes dynamic task allocation across edge-fog-cloud crowdsensing systems.” Scienmag, 3 September 2026, https://scienmag.com/new-framework-optimizes-dynamic-task-allocation-across-edge-fog-cloud-crowdsensing-systems/. Accessed 3 September 2026.
Denise Maddox. “New framework optimizes dynamic task allocation across edge-fog-cloud crowdsensing systems.” Scienmag. September 3, 2026. https://scienmag.com/new-framework-optimizes-dynamic-task-allocation-across-edge-fog-cloud-crowdsensing-systems/
Copy citation Download RIS
Tags: bio-inspired optimization frameworkbio-inspired optimization frameworkscost reduction in edge computingcrowdsensing deployment challengescrowdsensing task allocationdynamic task allocationdynamic task scheduling in IoTedge-fog-cloud computing hierarchyedge-fog-cloud computing optimizationenergy-efficient crowdsensingenergy-efficient crowdsensing networksInternet of Things data managementIoT sensor data collectionIoT sensor data processing efficiencyMobile crowdsensingmobile crowdsensing data collectionmobility-aware task distribution algorithmsMOFARO algorithmmulti-objective optimization for crowdsensingmulti-objective optimization in distributed systemsreal-time resource allocation in IoTreal-time sensor data processingscalable crowdsensing system managementscalable task scheduling
