Heuristics for improving trip-vehicle fitness in on-demand ride-sharing systems

On-demand ride-sharing services are emerging alternatives to classical transport modes. Combined with self-driving vehicles, this movement has potential to shape the future of our mobility. To make full use of the potential, such services need to be scalable with growing demand. Assigning real-time trip requests to vehicles such that the driving costs are minimized is computationally expensive, but has to be done fast. This work proposes an approach to reduce the processing time it takes to assign a trip request to a vehicle. The solution is a trip-vehicle fitness estimation framework that is flexible enough to utilize any fitness measure and is self-adjusting through feedback loops. We analyze the placement of a trip request within a vehicle schedule, present and implement three fitness measures. The resulting system is evaluated based on performance, customer satisfaction and vehicle costs criteria by running simulations. The evaluation results indicate significant pe rformance improvement and noticeable improvements in terms of customer satisfaction and vehicle costs.