Slope Perception from Monoscopic Field Images: Applications to Mobile Robot Navigation

摘要

When remotely navigating a mobile robot, operators must estimate the slope of local terrain in order to avoid areas that are too steep to climb or that slope so steeply downward that the operator would lose control of the rover. Although many rovers are equipped with sensor systems to aid the operator in this task, it is sometimes necessary to estimate slopes from two-dimensional images, either when planning operations or when the operator wishes to monitor the results of a sensor system. This experiment compares the operator's estimates of the slope in Martian terrain with the actual slope determined from three-dimensional data. The ten participants overestimated the slope of the indicated regions by an average of 19° (SD16°). An analytic model of the error, based on psychophysical analysis, accurately predicts the average magnitude of the errors. Implementation of this model eliminates an average amount of participant error. However, the large estimate variance within and between participants and images still poses a challenge for accurate slope estimation.