自駕公車強化學習派遣之成本效益分析

Costs and Benefits Analysis of Autonomous Bus Using Reinforcement Learning Dispatching

自駕公車在全世界超過七十幾個城市推廣測試，應用先進無人駕駛技術可以免除駕駛人力成本、降低交通事故，有很大的潛力為公共運輸營運與服務帶來革命性的改變。自駕公車要能成功的推廣，關鍵因素在於成本結構是否具有競爭力。考量自駕公車本質上是「去中央指派、獨立運作」的智慧體，其運作模式和成本結構與傳統公車系統有很大的不同。本研究於是發展自駕公車強化學習派遣的成本函數，並應用一個動態運量指派的公共運輸模擬系統，來分析自駕公車在往返公車路線上的營運總成本，包括業者營運成本、乘客等待時間成本與乘客車內時間成本。研究結果顯示自駕公車人工智慧派遣在社會成本最佳化狀況下，較現行固定班表普通公車具有明顯的成本效益。

More than 70 cities are having trials on autonomous buses in the past five years. With the advances in driverless technology, autonomous buses are able to eliminate crew cost and reduce traffic accidents hence allow to potentially revolutionize existing public transportation systems. A key determinant of autonomous buses viability is the competitiveness of their cost structures. Considering autonomous buses are naturally decentralized and self-governing intelligent agents, there have different operation model and cost structure than traditional bus system. This paper therefore proposes a cost function of autonomous bus based on a reinforcement learning dispatching method for autonomous bus operation. The total cost of autonomous bus, which is the sum of operating, waiting, riding cost in a line loop route were captured by using a dynamic public transportation assignment and operation simulation. The results show that autonomous buses with artificial intelligence dispatching has obvious cost-effectiveness compared with existing scheduled bus system while at the circumstance of optimizing social cost.