Abstract: Systems, methods, and non-transitory computer-readable media can construct a simulation framework for a ride sharing service. The simulation framework comprises a simulation environment and an agent comprising one or more algorithms including an order dispatching algorithm and a driver reposition algorithm. One or more states of the simulation environment include information about a plurality of drivers and a plurality of trip order requests, and can be provided to the agent. One or more actions from the agent can be obtained. The one or more actions comprises at least one of: a plurality of matches between the plurality of drivers and the plurality of trip order requests, or a plurality of reposition destinations for a subset of the plurality of drivers. The one or more states of the simulation environment can be updated based on the one or more actions.

Abstract: Deep reinforcement learning may be used for vehicle repositioning on mobility-on-demand platforms. Information may be obtained. The information may include a current location of a vehicle on a ride-sharing platform. A set of paths originated from the current location of the vehicle may be obtained. Each of the set of paths may have a length less than a preset maximum path length. A set of expected cumulative rewards along the set of paths may be obtained based on a trained deep value-network. A best path from the set of paths may be selected based on a heuristic tree search of the set of expected cumulative rewards. A next step along the best path may be recommended as a reposition action for the vehicle.

Abstract: Multi-objective distributional reinforcement learning may be applied to order dispatching on ride-hailing platforms. A set of historical driver trajectories and a set of driver-order pairs may be obtained. A weight vector between a first reward of the set of historical driver trajectories and a second reward of the set of historical driver trajectories may be determined using inverse reinforcement learning (IRL). A first value function and a second value function may be jointly learned using distributional reinforcement learning (DRL) based on the historical driver trajectories and the weight vector. A set of scores comprising a score of each driver-order pair in the set of driver-order pairs may be determined based on the weight vector, the first value function, and the second value function. A set of dispatch decisions may be determined based on the set of scores that maximizes a total reward of the set of dispatch decisions.