Robotaxis as Computing Clusters: A Stochastic Modeling Approach

In the future, passengers will likely be able to request an autonomous taxi or robotaxi to transport them to their destinations. The operators of a robotaxi fleet may want to operate their vehicles continuously, as these vehicles are often costly to build and operate. However, as the passenger service requests arrive randomly, there will be idling taxis. The fleet operators can utilize the computing resources of idling taxis to execute tasks from external customers. In this paper, we evaluate the task execution performance of a robotaxi fleet. We assume that passenger service has preemptive priority over task execution. Thus an idling taxi executing a task may have to serve a passenger. We study the system's performance under infinite and finite backlogs of tasks. In the infinite backlog case, there will always be tasks to execute for idling taxis. In this case, we derive the probability distribution of the number of tasks the fleet can serve during a cycle, which is the interval between two consecutive time points when the entire fleet becomes idle. In the finite backlog case, we assume the tasks requiring service arrive at the system according to a Poisson process and derive an approximation for the average task delay. Finally, we present the numerical results for the analysis and the simulation results to show the correctness of the work.