On improving the performance of a multi-agent taxi dispatch system

Abstract

The taxi dispatching problem has been a hot topic in recent years. All taxi operating companies are seeking to find efficient ways to dispatch taxi in response to customer requests.

Quite a lot of simulations have been done to investigate this problem. One recent project work called N-Taxi groUp Collaborative (NTuCab) by seniors before me applied a multi agent approach to concurrently assignment assign multiple requests to multiple taxis. The project has achieved much better performance than the current deployed system in terms of customer waiting time and taxi empty cruising time.

Aimed at making further progress in improving the efficiency of the multi-agent taxi dispatch system, further investigation is carried out to the NTuCab system in this project. This project focuses on relaxing one of the assumptions made by NTuCab during negotiation among taxi agents. The assumption was that all taxis should halt immediately when they are communicating with each other and continue to move only after the negotiation is concluded. To relax this assumption, a new dispatch policy called “Negotiation on the Go” is proposed, where, to allow taxis to be on the move during negotiation, instead of using the current taxi position, a local estimate of the taxi position right after negotiation is used to calculate the shortest time to reach the pick-up location of each request.

By implementing the new policy with the help of intelligent agents JADE platform and running the simulation experiments on MITSIMLab, a microscopic traffic simulator, simulation results are obtained and a detailed analysis is done based on the results. The simulation result and analysis show that with an appropriate choice of the Negotiation Timing Index, the new policy is able to achieve better performance for NTuCab.