task allocation

Task allocation (TA) for mobile robot swarms in hospital, logistics, and warehouse environments demands simultaneous optimization of energy consumption and robot count to enhance economic viability and environmental sustainability. Method: We conduct a systematic review of AI-driven TA approaches and propose the first sustainability-oriented multidimensional evaluation framework—covering centralized/distributed optimization, reinforcement learning, and multi-agent game-theoretic methods—validated via simulations in ROS, MATLAB/Simulink, and ARGoS. Performance is quantitatively benchmarked across energy consumption, task completion time, and scalability. Contribution/Results: We introduce the first TA evaluation framework explicitly targeting energy efficiency and resource minimization. Our analysis identifies critical research gaps and provides actionable algorithm selection guidelines. The framework delivers both theoretical foundations and practical implementation pathways for green, cost-effective deployment of mobile robots in industrial logistics.

Multi-robot task allocation faces three key challenges: task priority constraints, intra-task coordination, and coalition-based collaboration among heterogeneous robots. Method: This paper proposes an online iterative reallocation algorithm that jointly models task dependencies via a task dependency graph and captures coalition-scale efficiency gains through a coalition-size effect model. To address the underlying NP-hard optimization problem, the method employs a network-flow-based approximation, validated via mixed-integer programming and greedy heuristics. High-fidelity simulation integrates realistic robot dynamics and physics-based modeling. Contribution/Results: Compared to offline approaches, the algorithm significantly improves robustness against task failures and model uncertainties. Experimental evaluation—across both stochastic task settings and real-world mission scenarios—demonstrates superior plan quality, enabling effective modeling, dynamic replanning, and coordinated execution of complex multi-robot missions.

Task allocation in heterogeneous multi-robot systems operating under multiple concurrent tasks remains challenging due to dynamic task urgency, execution progress, and the need for coordination without centralized control. Method: This paper proposes a decentralized approach grounded in global game theory: task urgency and execution progress are modeled as global signals; robot utility functions are formulated as linear functions of these signals; and a mixed-strategy Nash equilibrium is computed directly via matrix inversion—bypassing iterative optimization or inter-robot communication. Contribution/Results: To our knowledge, this is the first application of global games to multi-robot task allocation. The method ensures load balancing while eliminating task conflicts, achieves O(n³) computational complexity, and demonstrates significant improvements in task completion rate and resource utilization in simulations. It offers high efficiency, scalability, and full decentralization—making it suitable for large-scale, dynamic robotic deployments.

Traditional multi-robot task allocation (MRTA) assumes each robot executes only one task at a time, ignoring the practical capability of modern robots to concurrently perform multiple subtasks—introducing complex physical interactions such as spatial conflicts and resource contention. Method: This work proposes a novel MRTA framework explicitly incorporating physical constraints into multi-task assignment. We formulate the problem as a constrained optimization and introduce two solution approaches: (i) an exact compilation method based on weighted MAX-SAT, and (ii) an efficient greedy heuristic algorithm. Contribution/Results: To our knowledge, this is the first framework unifying multi-tasking capability with physical feasibility in MRTA. Extensive evaluations—including simulated site-clearing scenarios and real-robot experiments—demonstrate a 37% improvement in task completion efficiency over single-task baselines. The approach significantly enhances system scalability and robustness in executing complex, concurrent tasks under realistic operational constraints.

This paper addresses the energy-aware task allocation problem for multimodal robot swarms capable of mode switching (e.g., flying, driving, walking). Methodologically, it introduces a joint task–execution-mode optimization framework: first modeling multimodal robots as dynamic graphs and designing graph encodings to jointly represent robot states, environmental constraints, and multimodal dynamics; then formulating a constrained nonlinear optimization model minimizing total energy consumption, accompanied by sufficient convergence conditions applicable to general kinematic and dynamic models. Contributions include: (1) simultaneous decision-making for task assignment and locomotion-mode selection; (2) guaranteed real-time performance and theoretical convergence; and (3) empirical validation—via simulations and representative scenarios—demonstrating reduced energy consumption and enhanced execution stability.

This paper addresses the dynamic deployment problem for multi-robot systems under communication connectivity constraints: robots must collaboratively reach points of interest (primary targets) to gather information and relay data reliably to a static base station via relay nodes; post-task, robots must be reassignable to new targets. We propose a two-stage optimization framework. In Stage I, we jointly generate a connected topology—incorporating both primary targets and relay positions—via clustering and tree-based modeling. In Stage II, a heuristic scheduling algorithm optimizes intra-cluster target assignment and visit sequencing. Our method explicitly enforces connectivity constraints, supports dynamic redeployment, and scales efficiently with system size. Experiments across varying robot counts and large-scale target sets demonstrate that the approach rapidly yields high-quality suboptimal solutions, significantly improving deployment efficiency and system scalability.

This work addresses the challenge of task allocation and scheduling for heterogeneous multi-robot systems in long-duration missions, where depletable resources such as battery capacity impose critical constraints. The paper introduces TRAITS, a novel framework that explicitly models the supply rate of depletable capabilities and integrates it directly into the task assignment process. By formulating a nonlinear program that jointly optimizes coalition composition and capability provisioning strategies, TRAITS generates efficient and feasible schedules that satisfy both temporal and energy constraints. In contrast to existing approaches, TRAITS maintains computational tractability while simultaneously accounting for complex capability requirements and resource consumption dynamics, thereby significantly improving both the feasibility and performance of mission execution.

This work addresses the tight coupling among robot design, formation control, and task planning in heterogeneous multi-robot systems, where existing approaches lack a unified framework that simultaneously accounts for task requirements and system-level trade-offs. To bridge this gap, the paper introduces a formal co-design framework grounded in monotone co-design theory, which abstracts robots, formations, and planners as interconnected modules with well-defined interfaces. This enables joint optimization under explicit task-performance constraints. The proposed framework is the first to support task-driven, composable, and scalable co-design, allowing seamless integration of new robot types, task configurations, and perception objectives while systematically discovering non-intuitive solutions with optimality guarantees. Case studies demonstrate that the approach significantly outperforms existing methods in flexibility, scalability, and interpretability.

This work addresses the online task allocation problem in dynamic multi-robot systems under task uncertainty, time-window constraints, and limited communication. The authors propose a distributed framework based on perceptual regions and communication graphs, introducing for the first time the game-theoretic iterative best response (IBR) mechanism to this setting. Each robot optimizes its marginal contribution using only local observations, enabling efficient task allocation in a decentralized manner. Experimental results in a simulated urban delivery scenario with up to one hundred drones demonstrate that IBR achieves task completion rates comparable to or better than those of EDD, the Hungarian algorithm, and SCoBA, under both full and sparse communication topologies, while incurring significantly lower computational overhead. The approach thus balances efficiency and scalability in large-scale multi-robot coordination.

This work addresses the dynamic robot resource allocation problem in heterogeneous multi-team systems by proposing an altruistic collaboration mechanism grounded in Hamilton’s rule from evolutionary ecology—a principle introduced here for the first time to multi-team resource allocation. The approach employs a graph neural network policy trained centrally but executed in a decentralized manner, jointly modeling heterogeneity in robot capabilities, transition costs, and interdependencies in team contributions to yield scalable, near-optimal allocation decisions. Evaluation in a fire-rescue simulation environment demonstrates that the method achieves near-optimal performance while effectively enabling efficient collaboration in large-scale, heterogeneous multi-agent systems.

This work addresses the challenge of jointly optimizing task assignment, scheduling, and path planning in multi-robot systems under spatiotemporal and resource constraints to maximize mission efficacy. The authors propose the STEAM problem framework, which—unlike prior approaches—models task efficacy as a continuous function and introduces a trait-efficacy mapping that captures the relationship between robot capabilities and task performance. To efficiently learn this mapping, they design a feasibility-aware active learning mechanism and develop E-ITAGS, an incremental task assignment graph search algorithm that simultaneously optimizes task efficacy and temporal constraints. Experimental results in emergency response scenarios demonstrate that E-ITAGS significantly improves task efficacy while strictly satisfying all constraints; furthermore, the active learning component exhibits high sample efficiency and comes with theoretical guarantees on suboptimality.