Optimizing Paratransit Routing Considering Dwell Time Uncertainty

This study presents a novel paratransit routing approach designed to address the challenges posed by dwell time uncertainty and the diverse context of requests in serving individuals with disabilities and the elderly. The complexity of this problem is further compounded by the dynamic and uncertain nature of traveler behavior, necessitating the use of efficient and adaptive learning techniques to yield effective solutions. Our proposed approach employs reinforcement learning techniques to minimize delays, while taking into account the uncertain nature of dwell times and the specific characteristics of each request. To illustrate the sequential update process, a numerical example is provided, showcasing the posterior distribution computations for predicted dwell time, as well as the updating of its mean and variance values. Through continuous adaptation and refinement of dwell time predictions, the model achieves more accurate and reliable estimates for unserved requests. The proposed framework makes a significant contribution to the optimization of paratransit routing by obtaining accurate and reliable estimates for unserved requests which can be a significant portion of service time for paratransit. By simultaneously reducing travel time and uncertainty in dwell time predictions, it offers the potential for enhancing the overall quality of paratransit service, resulting in improved transportation experiences for individuals with disabilities and the elderly.