Christopher A. Brooks

Vibration-based terrain classification for planetary exploration rovers

Safe, autonomous mobility in rough terrain is an important requirement for planetary exploration rovers. Knowledge of local terrain properties is critical to ensure a rovers safety on slopes and uneven surfaces. Visual features are often used to classify terrain; however, vision can be sensitive to lighting variations and other effects. This paper presents a method to classify terrain based on vibrations induced in the rover structure by wheel-terrain interaction during driving. This sensing mode is robust to lighting variations. Vibrations are measured using an accelerometer mounted on the rover structure. The classifier is trained using labeled vibration data during an offline learning phase. Linear discriminant analysis is used for online identification of terrain classes, such as sand, gravel, or clay. This approach has been experimentally validated on a laboratory testbed and on a four-wheeled rover in outdoor conditions.

Vibration-based Terrain Analysis for Mobile Robots

Safe, autonomous mobility in rough terrain is an important requirement for planetary exploration rovers. Knowledge of local terrain properties is critical to ensure a rover’s safety on slopes and uneven surfaces. This paper presents a method to classify terrain based on vibrations induced in the rover structure by wheel-terrain interaction during driving. Vibrations are measured using an accelerometer on the rover structure. The classifier is trained using labeled vibration data during an off-line learning phase. Linear discriminant analysis is used for on-line identification of terrain classes such as sand, gravel, or clay. This approach is experimentally validated on a laboratory testbed.

Self-supervised terrain classification for planetary surface exploration rovers

In future planetary exploration missions, improvements in autonomous rover mobility have the potential to increase scientific data return by providing safe access to geologically interesting sites that lie in rugged terrain, far from landing areas. To improve rover-based terrain sensing, this paper proposes a self-supervised learning framework that will enable a robotic system to learn to predict mechanical properties of distant terrain, based on measurements of mechanical properties of similar terrain that has been traversed previously. In this framework, a proprioceptive terrain classifier is used to distinguish terrain classes based on features derived from rover–terrain interaction, and labels from this classifier are used to train an exteroceptive (i.e., vision-based) terrain classifier. Once trained, the vision-based classifier is able to recognize similar terrain classes in stereo imagery. This paper presents two distinct proprioceptive classifiers—a novel approach based on optimization of a traction force model and a previously described approach based on wheel vibration—as well as a vision-based terrain classification approach suitable for environments with unexpected appearances. The high accuracy of the self-supervised learning framework and its supporting algorithms is demonstrated using experimental data from a four-wheeled robot in an outdoor Mars-analogue environment.

Self-Supervised Classification for Planetary Rover Terrain Sensing

Autonomous mobility in rough terrain is key to enabling increased science data return from planetary rover missions. Current terrain sensing and path planning approaches can be used to avoid geometric hazards, such as rocks and steep slopes, but are unable to remotely identify and avoid non-geometric hazards, such as loose sand in which a rover may become entrenched. This paper proposes a self-supervised classification approach to learning the visual appearance of terrain classes which relies on vibration-based sensing of wheel-terrain interaction to identify these terrain classes. Experimental results from a four-wheeled rover in Mars analog terrain demonstrate the potential for this approach.

Visual, tactile, and vibration-based terrain analysis for planetary rovers

Future planetary exploration missions can require rovers to perform difficult tasks in rough terrain, with limited human supervision. Knowledge of terrain physical characteristics would allow a rover to adapt its control and planning strategies to maximize its effectiveness. This paper describes recent and current work at MIT in the area of onboard terrain estimation and sensing utilizing visual, tactile, and vibrational feedback. A vision-based method for measuring wheel sinkage is described. A tactile method for on-line terrain parameter estimation is also presented. Finally, a method for terrain classification based on analysis of vibration in the rover suspension is described. It is shown through simulation and experimental results that these methods can lead to accurate and efficient understanding of a rovers physical surroundings.

Visual wheel sinkage measurement for planetary rover mobility characterization

Wheel sinkage is an important indicator of mobile robot mobility in natural outdoor terrains. This paper presents a vision-based method to measure the sinkage of a rigid robot wheel in rigid or deformable terrain. The method is based on detecting the difference in intensity between the wheel rim and the terrain. The method uses a single grayscale camera and is computationally efficient, making it suitable for systems with limited computational resources such as planetary rovers. Experimental results under various terrain and lighting conditions demonstrate the effectiveness and robustness of the algorithm.

Terrain classification and identification of tree stems using ground-based LiDAR

To operate autonomously in forested terrain, unmanned ground vehicles must be able to identify the load-bearing surface of the terrain (i.e., the ground) and obstacles in the environment. To travel long distances, they must be able to track their position even when the forest canopy obstructs GPS signals, e.g., by tracking progress relative to tree stems. This paper presents a novel, robust approach for modeling the ground plane and tree stems in forests from a single viewpoint using a lightweight LiDAR scanner. Ground plane identification is implemented using a two-stage approach. The first stage, a local height-based filter, discards most nonground points. The second stage, based on a support vector machine classifier, identifies which of the remaining points belong to the ground. Main tree stems are modeled as cylinders or cones to estimate the diameter 130 cm above the ground plane. To fit these models, candidate main stem data are selected by finding points approximately 130 cm above the ground. These points are clustered into separate point clouds for each stem. Cylinders and cones are fit to each point cloud, and heuristic filters identify which fits correspond to tree stems. Experimental results from five forested environments demonstrate the effectiveness of this approach. For ground plane estimation, the overall classification accuracy was 86.28% with a mean error for the ground height of approximately 4.7 cm. For stem estimation, up to 50% of the main stems were accurately modeled using cones, with a root mean square diameter error of 13.2 cm.© 2012 Wiley Periodicals, Inc.

Visual detection of novel terrain via two-class classification

Remote sensing of terrain characteristics is an important component for autonomous operation of mobile robots in natural terrain. Often this involves classification of terrain into one of a set of a priori known terrain classes. Situations can frequently arise, however, where an autonomous robot encounters a terrain class that does not belong to one of these known classes. This paper proposes an approach for visual detection of novel terrain based on a two-class support vector machine (SVM) for situations when known terrain classes can be confidently associated with only a subset of the training data. Experimental results from a four-wheeled mobile robot in Mars analog terrain demonstrate the effectiveness of this approach.

Ground plane identification using LIDAR in forested environments

To operate autonomously in forested environments, unmanned ground vehicles (UGVs) must be able to identify the load-bearing surface of the terrain (i.e. the ground). This paper presents a novel two-stage approach for identifying ground points from 3-D point clouds sensed using LIDAR. The first stage, a local height-based filter, discards most of the non-ground points. The second stage, based on a support vector machine (SVM) classifier, operates on a set of geometrically defined features to identify which of the remaining points belong to the ground. Experimental results from two forested environments demonstrate the effectiveness of this approach.