We consider a non-standardized industrial environment where a fleet of large and heterogeneous automated guided vehicles (AGVs) operates along a roadmap that captures the geometric and operational constraints of the facility, including narrow bidirectional corridors, intersections, and irregular layouts. The vehicles differ in size, kinematic behavior, and motion capabilities, and are required to perform a sequence of pick-up and drop-off operations within a shared workspace.
Preliminaries
This section outlines the problem statement, the key concepts used to model the environment, and the assumptions that ensure the feasibility of the proposed approach.
Problem Statement
The objective of the traffic management system is to generate coordinated motion plans that enable all vehicles to navigate safely and efficiently while completing their assigned tasks. The problem integrates path planning and fleet coordination into a unified framework that ensures collision-free and dynamically feasible motion, while simultaneously detecting and resolving deadlocks and minimizing overall traversal cost in terms of throughput and travel time. Safe operation must be guaranteed by avoiding both spatial and temporal conflicts, maintaining continuous traffic flow even under high-density traffic conditions.
Industrial environments are inherently dynamic, as tasks may be reassigned, temporary obstacles can appear, and operational priorities may change during execution. The coordination framework must therefore operate in real time, continuously adapting vehicle trajectories to the evolving system state, variations in task assignment or workspace configuration, and situations that may result in deadlocks such as blocked corridors or mutual agent waiting.
Assumptions
The environment of the industrial plant must satisfy the following conditions.
Regarding the AGVs, the following simplifying assumptions are taken into account:
- When operating on the roadmap, the AGVs can only wait at locations or move along segments.
- The Task Lists, meaning the lists of tasks to be executed for each AGV, are provided by the Task Manager.
- The AGVs follow a fixed path, calculated during task assignment, to perform a specific task.
Traffic Management Architecture
The overall workflow of the proposed strategy is based on a central control unit, referred to as the Traffic Manager (TM). The TM periodically receives updated information on the state of each AGV, including localization data and status indicators.
The TM also acquires the Task Lists provided by the Task Manager. These lists specify the tasks currently assigned to each vehicle, such as picking up or dropping off goods, navigating toward designated locations, or reaching a charging station when the battery level is low.
L-MAPF Coordinator
The L-MAPF coordinator is responsible for generating space-time-consistent trajectories for all active AGVs, ensuring collision-free motion while enforcing the fixed paths assigned to each vehicle and respecting the structural constraints of the industrial layout.
Unlike one-shot planning strategies, the L-MAPF coordinator operates in a rolling-horizon fashion. Instances, each consisting of a set of trajectories, are periodically recomputed so that the system can continuously adapt to new task assignments, evolving AGV states, and execution uncertainties.
A rigorous formulation of the coordination problem solved at each instance requires a precise specification of all quantities involved in trajectory generation. Relevant elements include:
- The current target position associated with each AGV.
- The fixed path along which coordination is performed.
- The roadmap subgraphs used during planning.
- The static and dynamic obstacles that represent environmental and traffic constraints.
Case Study
The solution is implemented in C# and integrated into the TecnoFerrari Supervisor software for AGV systems, which includes the Task Manager module.
All experiments are conducted on a notebook equipped with an Intel Core i7-1165G7 processor with a 2.80 GHz clock speed and 16 GB of DDR4-3200 RAM. The experiments are validated in a real industrial environment in collaboration with the company.
The supplementary video shows some of the tests conducted in the automated factory.
Findings
The results obtained from the analyses offer a unified picture of how the proposed system behaves across different operating conditions.
The proposed traffic management system effectively coordinates a heterogeneous AGV fleet on non-standardized industrial roadmaps with narrow bidirectional corridors and reduced maneuvering space, while also delivering consistent performance in more regular and spacious layouts.
Throughput grows with fleet size until each layout reaches the saturation point imposed by geometry and interaction density. Beyond that limit, additional AGVs no longer enhance productivity, although the L-MAPF coordinator continues to handle conflicts within real-time bounds except in the most congested configurations.
The allocation-horizon parameter exerts limited influence on system-level performance. A broad range of values yields comparable results, with short horizons working best in irregular layouts and longer horizons remaining advantageous in more structured environments.
The ablation study indicates that both the anytime mechanism and the EHC strategy play a critical role in complex layouts. Disabling either one delays conflict detection, increases deadlock occurrences, and degrades throughput, flow time, and management efficiency. The effect is weaker in large, regular environments but remains beneficial whenever interaction density rises.
The comparative evaluation shows that the proposed system surpasses the rule-based TM, the state-of-the-art TM, and the priority-based L-MAPF variant across all non-standardized layouts. Anticipation of conflicts, coordinated motion under high-density traffic, and reliable deadlock detection and resolution produce consistently higher throughput and lower travel times in geometrically constrained settings.
The convergence analysis confirms that the ABH-CBS solver approaches the optimal and complete CBS solution as the time horizon expands. Execution-time measurements further show that real-time constraints are met across all instances while preserving the ability to recover the full CBS solution when additional computation is available.
Acknowledgments
The authors gratefully acknowledge Gruppo TecnoFerrari S.p.A., and particularly Anna Bellodi, Riccardo Benedetti, and Andrea Bargi from the Software Development Office, for their support in the preparation of this paper.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, or publication of this article.
Funding
This work was supported by the Socially-acceptable Extended Reality Models and Systems (SERMAS) Project of the European Union’s Horizon Europe Research and Innovation Program, grant agreement number 101070351.
ORCID IDs
- Alessandro Bonetti: 0009-0008-2068-6165
- Silvia Proia: 0000-0002-3801-4745
- Simone Guidetti: 0009-0002-1446-1142
- Lorenzo Sabattini: 0000-0002-2734-5549
Frequently Asked Questions
What problem does the traffic management system address? It coordinates large, heterogeneous AGVs operating in non-standardized industrial environments with narrow corridors, intersections, and irregular layouts.
How does the L-MAPF coordinator adapt to changing conditions? It periodically recomputes trajectory instances in a rolling-horizon process that adapts to new tasks, changing AGV states, and execution uncertainties.
What happens when the fleet reaches a layout’s saturation point? Throughput stops increasing because of geometric and interaction-density limits, although the coordinator continues handling conflicts within real-time bounds in most configurations.
Which components are important in complex layouts? The anytime mechanism and the EHC strategy are both important because disabling either delays conflict detection, increases deadlocks, and reduces performance.