Résumé:
A dynamic ridesharing system enables users to find journeys in real time by connecting
drivers and passengers who share a similar itinerary. It optimizes vehicle use by reducing
the number of cars on the road and promoting the sharing of transportation costs.
In our work, we focus on the dynamic matching problem in a ridesharing system.
We proposed to solve the problem using reinforcement learning. The proposed solu-
tion takes into account spatio-temporal constraints, capacity constraints, waiting time
constraints and detour constraints. Our proposed model aims to minimize passenger wai-
ting time and detour time.
We verified our model using real data from New York City’s public cab dataset, as well
as using a simulator we developed to evaluate the performance of the approach.
The results obtained are promising, reinforcing the validity of our choices and successfully
demonstrating the effectiveness and robustness of our real-time approach.
