Both centralized and distributed methods for fleet management are feasible. There are considerable advantages and disadvantages to each, so the choice between the two will ultimately depend on your overall requirements, and which method is a better fit for those requirements.
Open Robotics is currently working on an open source platform-agnostic mobile robot fleet management+integration framework called the “Robotics Middleware Framework” (RMF). You can find some high-level documentation on the project here. The landing page for the source code is here. This work is still in the R&D phase, but it’s actively moving towards production.
The traffic management in RMF is designed to use a distributed approach, although it supports centralized planning (as a rule of thumb, it’s easier to centralize a distributed system than to distribute a centralized system). As explained here there are two major dimensions to the traffic management:
Prevention. There is a centralized schedule that all robots report their intended trajectories to, and which all the robots can read (asynchronously). Using this information, the robots can plan optimal routes for themselves that shouldn’t interfere with any of the other robots.
Negotiation. When a traffic conflict is identified despite the preventative measures, a negotiation will be started between the conflicting robots, and they’ll talk peer-to-peer to negotiate an optimal resolution to their conflict.
A design goal of RMF is to allow multiple unrelated proprietary fleet managers to be able to cooperate with each other without changing the way those fleet managers are implemented. We anticipate many of those fleet managers will take a centralized approach to their traffic management, so RMF supports the ability to have one fleet manager that schedules and negotiates on behalf of many robots simultaneously.
The traffic management design in RMF is probably more complex than what’s strictly needed for applications where all robots will always be the same type of platform with the same API on the same network. But it does work for those scenarios, and it carries the benefit that your system won’t be locked into those requirements, making it easier to add new platforms in the future or migrate to more complex setups.