Integrated Operation Model for Autonomous Mobility-on-Demand Fleet and Battery Swapping Station

Optimal Spatiotemporal Pricing for Autonomous Mobility-on-Demand Systems: A Decomposition and Dynamic Programming Approach

“…Although the problem is formulated as a Markov decision process (MDP), exact dynamic programming methods are not applicable because of the continuous and high-dimensional state and action spaces. In the subsequent section, we develop a model predictive control (MPC) framework to tackle (9) and propose an integrated relaxation, decomposition, and dynamic programming approach to establish a theoretical upper bound.…”

“…But fortunately, demand in the near future is usually predictable, enabling the operator to devise an optimal operation strategy ahead of time. Hence, the AMoD operation relies on accurate demand prediction, which makes MPC wellsuited for solving (9). On the one hand, MPC exploits the proposed network flow model to anticipate future system output and potential constraint violations.…”

“…At the beginning of time step t, the platform observes the current system state S t and forecasts the travel demand in the subsequent H time steps, denoted as Q t = [q ij,t ] i,j∈V,t∈H . Then the platform maximizes the total profit in the next H steps by solving (9), except that the summation over time in the objective is for the prediction horizon H rather than the entire operation horizon T . The problem remains complex because of the nonlinear system dynamics and the existence of constraints.…”

“…Related Work: There is a growing body of literature investigating AMoD systems. Topics of related research include assignment [3], rebalancing [4] [5], routing [6] [7], charging [8] [9], parking [10], etc. Particularly, Zhang and Pavone [11] characterize AMoD systems as a queuing network model and find the optimal rebalancing policy by solving a linear program.…”

“…Iglesias et al [12] tackle the routing and rebalancing problems for AMoD fleet using a Baskett-Chandy-Muntz-Palacios (BCMP) queuing network, which captures the passenger arrival process, traffic conditions, state of charge, and vehicle availability at stations. Ding et al [9] consider the integrated operations of AMoD systems and battery swapping stations, where the two systems interact with each other via the swapping demand and prices. They formulate an expended network flow model to capture the interdependence and confirm the benefit of this integration for both systems.…”