Reinforcement Learning Framework for Dispatch Systems

Grab has developed and researched advanced techniques for optimizing the dispatch system, which matches bookings with drivers. This includes exploring the application of Reinforcement Learning (RL) through frameworks like DispatchGym to improve matching efficiency and cost-effectiveness. Additionally, research has focused on improving the theoretical and practical aspects of the matching algorithm itself, particularly by optimizing the computation of the cost matrix in bipartite matching problems. DispatchGym provides a research framework for RL studies, integrating an RL optimization approach and a dispatch simulation using the Gymnasium API, enabling rapid exploration and iteration of RL applications. It addresses challenges in aligning reward functions with business goals and sensitivity to control levers, empowering data scientists to refine contextual bandit approaches for optimizing dispatch hyperparameters.