Purchase this article with an account.
Michael K. McBeath, Thomas G. Sugar, Abhay Paranjape, Igor Dolgov, Wei Wang; Human and robot ball catching on a Hill: Is the control geometry on the level or atilt?. Journal of Vision 2007;7(9):118. doi: 10.1167/7.9.118.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Past research supports that humans, animals, and robots navigate to intercept moving airborne or ground-based targets by using a small set of simple, angular control heuristics. Pursuers navigate to keep the target image moving at a constant optical speed and direction. For fly balls the tangent of the vertical optical angle increases at a constant rate, while for grounders the cotangent decreases at a constant rate. Past research typically assumes that these optical control functions are based on a universal, level, world-based geometry, but this may only be true with level terrains. The present study examines human and mobile-robot fielders intercepting ground balls rolling up or down hills. We tested if navigational behavior is more consistent with optical-angle functions that are based on level, world-coordinate geometry, or on a reference frame rotated parallel to hill-slope. Both human and robot fielders pursued ground balls on a 4° slope. Measurements were made using an 8-camera Vicon motion-capture system with 60 Hz, 1 cm position resolution. Experiment conditions followed a 2-by-2 design with both uphill and downhill directions of ball-movement and locomotion. In the human experiment, three skilled fielders caught five grounders in each of the four conditions. The findings confirmed the optical control model was significantly more linear when based on a reference frame tilted with the hill slope. Results of the robot study exhibited the same pattern, with the robot catastrophically failing when programmed to use a level-geometry control mechanism in the ball-rolling-down, fielder-running-up condition in which humans also had the least variance explained. Overall, the tilted geometry model accounted for an average of over 98% variance in the vertical optical-angle function. The findings support that interceptive human action utilizes a geometry recalibrated to tilt with hill-slope, and that robotic control models are greatly improved by adopting this same strategy.
This PDF is available to Subscribers Only