A Traffic Management System for Large and Heterogeneous Vehicles in Narrow Industrial Environments

In the Logistics 4.0 era, the AGVs inside a factory or warehouse are evolving from a few vehicles on fixed routes into fleets of dozens or hundreds — different sizes, different capabilities, sharing narrow aisles. Once density rises, the problem stops being single-vehicle navigation and becomes fleet traffic management: who goes first, who waits, who reroutes. The traditional answer is negotiation-based priority assignment, with vehicles talking to each other to settle right-of-way. That works at low density, but in a dense industrial site with narrow corridors the negotiation itself becomes both the bottleneck and a source of gridlock.

Core Innovation

The paper formulates fleet traffic management as a lifelong multi-agent path finding (L-MAPF) problem and builds a complete system aimed at real industrial sites.

• NURBS roadmaps. The drivable network is not an imposed grid but is generated from non-uniform rational B-spline (NURBS) curves, so it can follow the curved, asymmetric, non-standardized free space of a real facility while guaranteeing locally optimal coordination and safe operation of large, heterogeneous vehicles.
• Bounded-horizon CBS inside rolling-horizon resolution. A modified bounded-horizon conflict-based search (CBS) is embedded in a rolling-horizon conflict resolution strategy, with an extended time horizon per agent so conflicts can be resolved effectively inside corridors identified by a topological map.
• Anytime resolution with adaptive horizons. Under compute pressure the system can return a usable solution immediately and keep improving it over time.
• An execution layer plus deadlock handling. The framework includes a safety- and standard-compliant execution layer for real AGVs and an advanced deadlock detection and resolution mechanism.

The positioning is explicit: unlike grid-based warehouse methods, this targets non-grid, narrow, bidirectional corridors at high traffic density, where AGVs of different sizes and capabilities run simultaneously.

Results

• Experiments in realistic industrial environments report throughput improvements of up to 11%.
• Comparisons span a conventional rule-based traffic management system, a state-of-the-art industrial method, and a priority-based L-MAPF variant.
• The system sustains higher throughput while maintaining continuous operation and improved efficiency.

Limitations

The abstract omits fleet size, corridor geometry, experiment duration, and the number of sites, so it is unclear over what scale the 11% gain holds. NURBS roadmaps must be built in advance, and the system’s adaptability to facility changes or temporary obstructions is not discussed. The standards behind the safety- and compliance-oriented execution layer, and any certification status, are not specified. The approach also assumes a structured industrial site; open settings with human workers or non-cooperative vehicles are outside its scope.

Industry Implications

For AGV fleet operators the payoff is directly financial: moving more material with the same vehicles is the easiest improvement to justify. The more telling signal is the deployment context — non-grid layouts, narrow aisles, and heterogeneous fleets are the normal state of real factories, while a great deal of academic work still assumes idealized grids. By packaging conflict resolution, deadlock handling, and a safety-compliant execution layer into one deployable system, this work sits visibly closer to the production line than to the benchmark.

---