Mobile robots have been successfully applied to many field applications, such as mining, planetary exploration, agriculture, and environmental monitoring, to name a few. In all these applications, robots face environments with physical characteristics that are a priori unknown and can heavily affect performance. In the case of ground robots, terrain roughness can affect the ability of a robot to navigate and even cause damage to its on-board hardware due to excessive vibration. To aid in these problems, methods to enable the robot to automatically learn terrain properties from its sensory data have been presented in the literature. However, such methods usually assume that localisation is accurate enough, without dealing with its inherent uncertainty.
展开▼