SUMOoD (SUMO on Demand) is a TraCI-based implementation of the simulation presented in Ronald et al..

The system deals with ad-hoc requests from passengers wanting to travel between two locations.

Current work has focused on comparing results with the original implementation in Delphi, using Axelrod’s criteria for replication. We have found that relationally the models are similar, however looking at distributional results there are differences, mostly due to the differences in the routing algorithms in both simulations.

We found some issues with moving from a network-based simulation to a traffic microsimulation:

  • Stops (picksups/dropoffs) could no longer take place at nodes in the network (that is, right in the middle of an intersection); as a result one link out of an intersection was chosen to represent a particular node;
  • The side of the road matters, which is actually an operational issue — some passengers cannot be expected to cross the road;
  • The speed along a link has an effect on the estimated travel and arrival time; a simplified model assuming maximum speed at all times is not overly realistic, however using a microsimulation to determine a realistic average speed is useful;
  • The details of the DRT scheme are crucial; should vehicles (with no passengers and no requests to service) park where they last dropped off a passenger, park at another specific parking place, choose the best place depending on where they predict the next request will come from, or return to the depot? On-demand vehicle congestion is a problem in SUMO (e.g., if many vehicles park at the same point at the same time, however this could be resolved with smarter optimisation/vehicle planning).

The code for SUMOoD can be found on GitHub. We are happy to hear of any potential improvements, and in particular are interested to hear from ME/MIT/MSc and bachelor students at our partner universities (Melbourne and Monash) who are interested in developing this further as a final-year project.