Optimized responsive public transport system

Software	Any powerful object oriented language
Covers	Discrete event simulation, AI planning, AI optimization techniques
Skills Required	Programming, some interest in artificial intelligence
Challenge	Conceptual ??? Technical ??? Programming ????

Brief Description

In the future, there will be increasing pressure to have fewer cars on the road, parking spaces will be harder to find as car parks are built on and local governments put pressure on employers to put pressure on employees not to drive to work (as any DMU staff member will tell you), and fossil fuels will become scarcer and more expensive. But the need for motorized transport will not decrease. Current bus services are inadequate for most people, inflexible, expensive, and are often a very inefficient use of the bus companies' resources; while taxi services are also inefficient in a different way with many taxis empty and few taxis shared.

Further in the future, we may see strong pressure to move to driverless robotic vehicles operating with a combination of autonomous local control, electronic interaction with other vehicles, and centralized coordinated route planning; as well as strong pressure to move to electric vehicles with limited ranges.

How can you use artificial intelligence techniques to ensure that people get their transportation needs met by a public transport system that is flexible enough to accommodate real need for door-to-door transportation, fast enough to meet the needs of regular commuters, and efficient enough not to destroy the planet?

Your challenge is to build a transport network simulator in which people make requests for journeys between different locations, and the coordinated public transport system moves them from one point to another. The journeys may involve transfers between different vehicles and may involve sharing vehicles as people are picked up or dropped off in sequence.

A road network. The simulator will need to build a road network of routes between different points, where each segment has a journey time. Ideally, this should be randomly generated, and parameterized so that road networks with different characteristics can be generated.
A discrete event simulator. The simulator will need to generate journeys between different points. Ideally these should not be evenly distributed but reflect the existence of peak travel times and popular destinations like shopping centres, schools and hospitals. The simulator will need to include the collection and drop-off of passengers and the movement of vehicles between nodes in the road network.
A scheduler. The simulator will need to determine which vehicles are allocated to which jobs, and whether vehicles can be shared between passengers, or whether passengers should be moved between vehicles, for example to put passengers onto larger buses for longer journeys between distant nodes (such as other towns). Constructing a coordinated plan might be done using a genetic algorithm.

Extensions

The road network should start simple, but can be made as complicated and realistic as you like. Similiarly the parameters governing the generation of journeys can be made more sophisticated, especially if the simulator treats passengers as entities distinct from journeys.

The project could compare performance given different sets of vehicles and different rules governing what the vehicles do. Is it actually possible to do better than a set of single passenger journey taxis or a rigidly timetabled set of buses?

The simulator could include costs for providing services and fares charged to passengers, and include price-inconvenience tradeoffs for particular journeys. (You would be a lot more willing to pay for a taxi if you have two heavy suitcases...)

Back to