In online food delivery, which is expanding all over the world, the increasing number of traffic accidentsduring delivery is becoming problematic. To ensure the safety of delivery workers and the residents, it isnecessary to understand the incentives of workers for behavior choices. While many existing sharingplatforms are pulling workers into the online labor process by such incentives as on-peak/off-peaksurcharges, workers try to accomplish more than their goals within a limited time. In this paper, to preventover-speeding during delivery, we compare and evaluate some task assignment methods and incentiveschemes by multi-agent simulation. We model worker's rational choices of behaviors based onreinforcement learning considering the profit and over-speeding of delivery workers.
