RAMBLE: ICT support for On-Demand Public Transport
Public transport operators traditionally offer scheduled services for their bus passengers. On demand provision has lagged behind, even though it is can be more convenient for passengers. For the operator, such on-demand services could lead to inefficient use of drivers and buses and might conflict with service commitments to serve disadvantaged groups. The RAMBLE project aimed to support the introduction of on demand public transport by improving the ability of its operators to devise efficient schedules assigning buses and their drivers to meet dynamic demand, and to take account of evolving traffic conditions while doing so.
The work was a targeted project under CONNECT between RouteMatch Inc (a provider of value added services to public transit operators), WIT and UCC. UCC focused on the online/dynamic scheduling aspects. WIT focused on route-finding and predicting demand.
- The project ran over 2 years (2018-2019). In the first year (Phase 1), WIT/TSSG developed a performance testbed with which to test a variety of routing engines. Service times and resource requirements were compared under a variety of experimental conditions and a set of recommendations was made. This resulted in an invention disclosure for the underlying testbed and its operating procedures, which is currently being reviewed by lawyers.
- During Phase 2, WIT/TSSG developed a new model, based on demand profiles, to represent time-varying demand for trips, given historical trip request data. The model is very flexible and, by taking careful account of probability, is also able to provide estimates of its uncertainty. The model is supported by data preprocessing, visualisation and supporting services, so that it offers potential uses for operational and strategic planning that go way beyond its intended use supporting the on-demand scheduler. This work is more evidence of TSSG’s use of advanced AI and machine learning techniques in applications, in this case in the transport domain.
For WIT/TSSG, Phase 1 of the project developed a performance testbed in the form of a set of server configurations, with a means of running performance experiments on this infrastructure. This was supplemented with machine learning to interpret the resulting performance and resource usage data and to recommend suitable routing engine configurations.
In Phase 2, we reviewed the existing models for demand prediction and found them wanting for this scenario. We developed a new model, drawing upon model designs from other application domains, then transformed the trip request data so that it suited our model. We supplemented the model with tooling to validate the results and to visualise the results.
During Phase 1, the project team comprised Bernard Butler. During Phase 2, the project team comprised Genaro Longoria and Bernard Butler.
- determine the best conditions and configuration settings for a variety of third-party routing engines, so that the scheduler could assemble candidate routes as quickly as possible
- ensure the routing engines themselves have enough ICT resources, but not too many, to reduce their operating costs
- the evaluation should take account of the need for dynamic updates of the route maps, e.g., due to road congestion or closures
- predict demand (over space and time) for on-demand personalised trip requests
- accept trip requests in a variety of formats and convert them to a standard format
- provide a means for users to query the model and to present predicted trip requests in an actionable format
- Two invention disclosures (one each for Phase 1 and Phase 2)
- The first of these is currently being reviewed by patent lawyers
- The industry partner is currently evaluating the Demand prediction outputs under the terms of an Evaluation Licence formulated by WIT’s TTO
Dr Bernard Butler, +353 51 84 5695
SFI through CONNECT, project code 13/RC/2077