Mobile robots extend the reach of exploration in environments unsuitable, or unreachable, by humans. Far-reaching environments, such as the south lunar pole, exhibit lighting conditions that are challenging for optical imagery required for mobile robot navigation.
Terrain conditions also impact the operation of mobile robots; distinguishing terrain types prior to physical contact can improve hazard avoidance.
This thesis presents the conclusions of a trade-off that uses the results from two studies related to operating mobile robots at the lunar south pole. The lunar south pole presents engineering design challenges for both tele-operation and lidar-based autonomous navigation in the context of a near-term, low-cost, short-duration lunar prospecting mission. The conclusion is that direct-drive tele-operation may result in improved science data return.
The first study is on demonstrating lidar reflectance intensity, and near-infrared spectroscopy, can improve terrain classification over optical imagery alone. Two classification techniques, Naive Bayes and multi-class SVM, were compared for classification errors. Eight terrain types, including aggregate, loose sand and compacted sand, are classified using wavelet-transformed optical images, and statistical values of lidar reflectance intensity.
The addition of lidar reflectance intensity was shown to reduce classification errors for both classifiers. Four types of aggregate material are classified using statistical values of spectral reflectance. The addition of spectral reflectance was shown to reduce classification errors for both classifiers.
The second study is on human performance in tele-operating a mobile robot over time-delay and in lighting conditions analogous to the south lunar pole. Round-trip time delay between operator and mobile robot leads to an increase in time to turn the mobile robot around obstacles or corners as operators tend to implement a `wait and see’ approach.
A study on completion time for a cornering task through varying corridor widths shows that time-delayed performance fits a previously established cornering law, and that varying lighting conditions did not adversely affect human performance. The results of the cornering law are interpreted to quantify the additional time required to negotiate a corner under differing conditions, and this increase in time can be interpreted to be predictive when operating a mobile robot through a driving circuit.
Source: The University of Western Ontario
Authors: Matthew Cross