Reliable dynamic in-vehicle navigation

Having started off from luxury makes and models, in-vehicle navigation systems are now gradually spreading through the entire vehicle fleet, as drivers appreciate their usefulness. Increasingly sophisticated systems are being developed, having much more advanced functions than simple driving directions. This thesis presents a new approach for in-vehicle navigation, in which travel time reliability is incorporated in the route finding component of the navigation system. Based on historical traffic data and in the absence of current traffic information, positions in the road network at which it is likely to encounter delays, are predicted and avoided as much as possible by the route finding algorithm.The thesis starts by reviewing shortest path algorithms and conjectures that the most appropriate algorithm to use is A*, which forms a vital part of the approach developed. Performing multiple runs of A* forwards and backwards on the road network, efficiency of the route finding procedure is achieved. The time-dependent version of the algorithm is also derived. Then, the thesis goes on to define reliability on a single link of the road network as the maximum delay that can be encountered with 90% confidence and extends this definition to derive the reliability of entire routes.Having introduced the route finding procedure and the concept of reliability, the thesis presents the in-vehicle navigation approach, which involves computing a more reliable route from the driver's origin to his/her destination than the fastest, if this is unreliable. Additionally, the approach aims at computing multiple alternative partially disjoint but equivalently reliable routes to the driver, such that the congestion feedback effect can be avoided as much as possible, without the need of carrying out a dynamic traffic assignment, which would be impracticable in an in-vehicle system. A number of constraints are introduced so as to ensure that the resulting routes are acceptable to the driver (are not too long, etc). Hence, the main concept lies in initially computing the fastest time-dependent route, then applying penalties to the links characterised as unreliable (increasing the link weights in inverse proportion to their reliability) and re-running the route finding algorithm so as to find a more reliable route. After each run, the route obtained is checked against the constraints and if it does not satisfy them, it is discarded, the penalties are reduced and a new route is sought. In order to obtain alternative partially disjoint routes, penalties are also applied to links that are already included in a previously computed and accepted route. The new algorithm, RDIN, is thus presented and mathematically formulated. An extension to RDIN for re-routing, RDIN-R, is also developed.The software tool developed for the application of RDIN and RDIN-R, the Adaptive Reliable Imperial Advanced Navigation Engine (ARIAdNE) is described. A simulation example is given for demonstration and preliminary validation; then a number of field experiments are carried out in Central London to test the method in a real road network environment and to compare its performance with an existing conventional car navigation system. The results suggest that the method is workable and precise, while at the same time it is a promising step forward in the field of in-vehicle navigation.