Evaluation and Comparison of Terrain Classification Techniques from LADAR Data for Autonomous Navigation

Autonomous navigation in natural environments remains a considerable challenge, primarily because of the difficulty in describing the environment of the robot in a way that captures the variability of natural environments Error! Reference source not found.. While it is possible – although by no means trivial – to build models of smooth, 3-D terrain, it is much more difficult to deal with areas that cannot be described by piecewise smooth surfaces. For example, the geometry of terrain types such as tall grass, densely forested areas, bushes are naturally described as 3-D texture rather than by smooth surfaces. Similarly, small obstacles such as wires or poles are hard to extract at a sufficiently long range from the robot for the planner to be able to properly anticipate the obstacles.