Robust Autonomous Navigation and World Representation in Outdoor Environments

Reliable localisation is an essential component of any autonomous vehicle system. The basic navigation loop is based on dead reckoning sensors that predict high frequency vehicle manoeuvres and low frequency absolute sensors that bound positioning errors. The problem of localisation given a map of the environment or estimating the map knowing the vehicle position has been addressed and solved using a number of different approaches. A related problem is when neither, the map nor the vehicle position is known. In this case the vehicle, with known kinematics, starts in an unknown location in an unknown environment and proceeds to incrementally build a navigation map of the environment while simultaneously using this map to update its location. In this problem, vehicle and map estimates are highly correlated and cannot be obtained independently of one another. This problem is usually known as Simultaneous Localisation and Map Building (SLAM). As an incremental algorithm, the SLAM in large outdoor environments must address several particular problems: the perception of the environment and the nature of features searched or observables with the available sensors, the number of features needed to successfully localise, the type of representation used for the features, a real time management of the map and the fusion algorithm, the consistency of the SLAM process and the data association between features mapped and observations. A good insight into the SLAM problem can be found in Durrant-Whyte & Bailey (2006). This chapter presents recent contributions in the areas of perception, representation and data fusion, focusing on solutions that address the real time problem in large outdoor environments. Topics such as DenseSLAM, Robust Navigation and non-Gaussian Observations in SLAM are summarised and illustrated with real outdoor tests.