For efficient and safe locomotion the gaits of legged robots should vary with the type of terrain. Hence, terrain surface classification is an important problem for this class of mobile robots. Prior research has developed approaches to proprioceptive terrain classification for both wheeled and limbed robots that use sensor measurements dependent upon the dynamics of the robot, which ultimately requires the classification system to be trained at a large number of operating conditions (e.g., vehicle speeds and loads). This research develops an approach to terrain identification based on pressure images generated through direct surface contact using a robot skin constructed around a high-resolution pressure sensing array. Terrain signatures for classification are formulated from the magnitude frequency responses of the pressure images. The methodology is used to train and test a classifier using dynamically measured pressure images from a one-legged hopping robot. Experimental tests yield high classification accuracies, which are independent with respect to changing robot dynamics (i.e., different leg gaits). The findings of this paper suggest the methodology can be extended to autonomous field robots, providing the robot with crucial information about the environment that can be used to aid stability over rough terrains and enhance motion planning over varying terrains.

Terrain classification is key to using terrain dependent control modes to improve performance of autonomous ground vehicles (AGVs). One of the most viable forms of terrain classification, reaction-based terrain classification, is subject to the problem of speed and load dependency, which requires collecting large data sets for algorithm training. The research presented here presents a method of interpolating point clouds called Singular Value Decomposition Interpolation or SVDI, which uses singular value decomposition, matrix logarithms and Catmull-Rom splines. The estimated point clouds can then substitute for empirical training data, thereby reducing the need to collect large data sets for algorithm training. Here, SVDI is applied to the problem of speed dependency using a mobile robot. Although it is seen that interpolated point clouds are not as effective as real data, interpolated point clouds are seen to be more effective than known point clouds that do not correspond to the desired vehicle speed. Therefore it is concluded that SVDI can effectively reduce the speed and load dependence of reaction-based terrain classification.

The need for terrain-dependent control systems on AGVs is evident when considering the variety of outdoor terrains many AGVs encounter. Although the idea of using terrain classification algorithms to identify the terrain and then update the control modes is well-established, the problem of how to intelligently update the control modes based on classifications has been left relatively unaddressed. This paper presents a simplistic rule, called the update rule, which decides when to change control modes based on past and present terrain classifications and is tuned using empirical data. Using experimental data from the experimental Unmanned Vehicle (XUV) mobile robot, this update rule is shown here to be both robust to misclassiflcations as well as sensitive to terrain transitions. This paper also develops and implements a sliding horizon approach to reaction-based terrain classification for improved sensitivity to terrain transitions. The update rule structure presented here is applicable to reaction and vision based terrain classification of individual terrains.

Terrain variations can greatly influence autonomous ground vehicle (AGV) performance. However, if the terrain is properly identified, the AGV control systems can be adjusted to better suit the terrain. Current terrain classification techniques are largely based on vision and terrain dependent vehicle reactions. But at this time both methods have limitations. Visual methods can easily be fooled by fallen leaves or other superficial ground coverings, whereas response-based methods yield poor results when the AGV speed is significantly different from the training speeds. In addition, current response-based methods are not capable of identifying the multiple terrains that a vehicle may simultaneously traverse, e.g., when one subset of wheels is on-road and one is off-road. This paper proposes a response-based method called terrain input classification to solve the issue of speed dependency and multiple terrain traversal. This method is dependent upon knowing the vibration transfer function of the AGV. The method has been verified through terrain-based simulation results and is ready for experimental testing.