robotics//fleet
A fleet is a group of robots or vehicles, from a few to several hundred and often of different types, that execute a common plan under a coordinator: the mobile robots of a warehouse, the guided vehicles of a factory floor, a team of inspection drones, the haul trucks of a mine. It is the form multi-robot systems take in production, and the coordinator is what makes the group a fleet.
A fleet is a group of robots or vehicles, from a few to several hundred and often of different types, that execute a common plan under a coordinator: the mobile robots of a warehouse, the guided vehicles of a factory floor, a team of inspection drones, the haul trucks of a mine. It is the form multi-robot systems take in production, and the coordinator is what makes the group a fleet.
The word is used against swarm, many simple and nearly identical agents that talk only to their neighbours and have no coordinator (swarm robotics). Almost everything that works in production is a fleet; almost everything called a swarm in videos is research, or a fleet in disguise. Warehouse systems such as Kiva Systems, now Amazon Robotics, move hundreds of robots with a central planner, fiducial markers on the floor that each robot reads to locate itself, and traffic rules on a grid. Factory AGV and AMR fleets (automated guided vehicles on fixed paths, autonomous mobile robots that plan their own) run under a central fleet manager, and VDA 5050, a standard of the German automotive and machinery associations VDA and VDMA, defines the interface between a fleet manager and vehicles from different makers so that one manager can drive them all.
Most fleet questions reduce to one: what to decide at the centre and what to let each robot decide.
A centre sees everything, can be optimal, is debugged from one log and can be certified; robots that decide locally survive the loss of the link and react faster than a round trip. The default answer and its exceptions are fleet architecture.
Talking costs. Range, bandwidth, packet loss and latency are the four field limits, and they create errors a lone robot never has (stale positions, two robots chasing the same task, a fleet split in two), all described in communication constraints.
Work has to be shared out. With a server, an assignment problem solver or a greedy rule is enough; without one, agents bid for tasks in auctions (task allocation).
Moving together (agreeing on a point, holding a formation, following a leader) is multi-agent control, built on the communication graph.
Behind a serious fleet sits replicated state (orders, tasks, maps) on several servers, the territory of distributed systems. A large fleet is also a complex system in its own right: a congested aisle slows robots, which keeps more of them in the aisle, and queues at the chargers decide throughput.
What each vehicle logs for maintenance and learning is fleet data.