The fleet-level tasks aim to work at a higher level of abstraction than the previous embodied robotic tasks LF, LFV, NAVV. Crucially instead of focusing on single robots, now a fleet of robots should be controlled.
Picture a city with several already autonomous vehicles. The question now posed at the fleet-level is how to best control vehicles to serve customers.
The actual social tasks will be described in more detail in “Autonomous Mobility-on-Demand”. Note that the sequence of tasks was chosen to gradually increase the difficulty of tasks by extending previous task solutions to more general situations.
Fleet management is a field of research which is recently has received increase attention. This section briefly and non-exhaustively presents related work to the fleet management and AMoDeus challenge.
In  several autonomous mobility-on-demand control algorithms were compared to a simulation scenario for the city of Zurich, Switzerland. It was shown
that a simple load-balancing heuristic reaches peak mean waiting times of 5 min with 10’000
vehicles, while more advanced control policies achieve the same performance with less than 9’000 vehicles.
A case study for Berlin presented in  simulates city-wide replacement of transportation systems with robotic taxis in Berlin. The results are obtained with a heuristic algorithm and for a large number of agents.
In  the authors use analytic results to estimate who many robotic taxis could satisfy the entire transportation demand of the country of Singapore. They estimate that $\approx 300,000$ taxis would be sufficient to serve demand with acceptable waiting times.
In  the relation of AMoD systems and queueing systems is detailed and a method to compute the fleet sizes required for certain levels of performance is presented.