Multi-sensor Estimation for Mobile Robot Rough-Terrain Traversability

Ground autonomous vehicles have important potential applications, such as reconnaissance, patrol, planetary exploration and military applications. To accomplish tasks on rough-terrain, control and planning methods must consider the physical characteristics of the vehicle and of its environment. Failure to understand these properties could lead to vehicle endangerment and mission failure. This paper describes recent and current work at Mobile Robotics Laboratory of the Politecnico of Bari in the area of rough-terrain traversability and sensing. Our research aims at studying vehicle-terrain interaction to provide reliable methods for predicting mobile robot tractive effort and assessing whether a given terrain can be safely traversed, which is usually referred to as traversability. An experimental framework for terrain characterization and identification is presented which employs a cylindrical shaped mobile robot as a tactile sensor. Vision-based methods are introduced for estimating vehicle kinematics and sinkage. Visual measurements are as accurate as conventional methods with the additional advantage of being contact-free. It is shown that the proposed framework is effective in experimental trials and can lead to efficient understanding of mobile robot physical surroundings. Such methods could enhance rough-terrain traversability through integration with control and planning algorithms.