Multi-task dispatch of shared autonomous electric vehicles for Mobility-on-Demand services - combination of deep...

Area of Science:

• Intelligent Transportation Systems
• Operations Research
• Artificial Intelligence

Background:

• Autonomous Mobility-on-Demand (AMOD) systems offer sustainable urban transport.
• Efficient fleet management is crucial for shared autonomous electric vehicles (SAEVs).
• Existing systems face challenges in optimizing recharging, delivery, and repositioning tasks.

Purpose of the Study:

• To develop an optimal task assignment policy for SAEVs.
• To enhance the operational efficiency and user satisfaction of AMOD systems.
• To formulate and solve the fleet dynamic operating process as a multi-agent multi-task dynamic dispatching problem.

Main Methods:

• Formulation as a Markov Decision Process (MDP).
• Recharging and delivery tasks modeled as maximum weight matching problems (Kuhn-Munkres Algorithm).
• Repositioning tasks modeled as a maximum flow problem (Edmond-Karp Algorithm).
• Integration of a novel reward function and a Back Propagation-Deep Neural Network for state-value estimation.

Main Results:

• A reward function balancing income and satisfaction increased revenue by 33.2%.
• Task allocation repositioning boosted total revenue by 50.0%.
• An improved state-value function led to a 2.8% revenue increase.
• Combined charging and repositioning strategies significantly reduced user waiting times and improved satisfaction.

Conclusions:

• The proposed MDP-based approach effectively optimizes SAEV fleet operations.
• Balancing economic factors with user experience is key to AMOD success.
• Advanced algorithms and AI enhance the performance and user-centricity of autonomous mobility services.