The DARP problem, also known as the Dial-a-Ride Problem, is a type of transportation problem that involves finding the most efficient way to pick up and drop off a group of passengers at different locations. The objective is to minimize the total distance traveled while satisfying the constraints of the problem, such as the capacity of the vehicles and the pickup and drop-off times for each passenger. Traditionally the problem is formulated as mixed-integer-programming task, and solved by the branch and cut algorithm. In this project we investigated a possible, online turn-based formulation and tried to achieve feasible DARP solutions using supervised machine learning techniques and reinforcement learning, using a Transformer Encoder neural network.
Clone the repository and setup conda environment.
git clone git@github.com:hbenedek/dial-a-ride.git && cd dial-a-ride
conda env create --prefix ./env --file environment.yml
conda activate ./env