• Sonia.NRI152

Kod*lab Menu

Internal Links (Login Required)

2015 NSF National Robotics Initiative PI Meeting, 2015.

Sonia Roberts and D. E. Koditschek
Electrical and Systems Engineering, University of Pennsylvania


Desertification is an increasingly severe problem in the southwest United States and around the world. Datasets with many repetitions of experiments close to each other in space and time, especially in response to a weather event, could lead to new scientific models of dune formation and movement. RHex performs well on complex terrain and offers a “plug-and-play” interface for scientists interested in taking data from multiple sensors simultaneously, offering a possible avenue to collecting this new type of dataset.

Locomotion on sand dunes is difficult and unpredictable. We tested RHex’s locomotion performance in three deserts with dunes sloped up to 29 degrees, thin dune crests, and occasional bushes. While RHex could access every location on a dune when allowed to use an “easy” path up the lower-inclination crest, performance was unpredictable on steep inclinations. For example, we predicted that wider legs would always perform better, but found that while this was true on flat interstitial areas, on sufficiently high slopes they performed worse. Since it is much slower to access the top of a dune via the crest than the steeper slip or windward faces, and since specific value to the underlying aeolian science may likely accrue to data collected along specific transects, we are motivated to understand the unpredictable performance to improve locomotion and avoid predictable robot failures.

Experimental physics hints at a method to interpret locomotion performance. Recent robophysics experiments revealed a scale-invariant empirical pattern predicting a dimensionless forward speed from the percent of the leg penetrating the sand. These experiments assumed rigid legs and flat, uniform granular media preparations, and the environments and robots were carefully controlled. However, our application uses compliant legs, a mix of ground inclinations, and no control of factors such as the volume fraction or particle homogeneity of the natural sand.

Extension to more realistic conditions produces a promising field assessment. We verified these results in a far less controlled environment to (1) account for leg compliance by allowing the leg length to vary; (2) predict current draw using only video from one high-speed camera to measure the leg penetration ratio; and (3) predict performance on inclines using leg penetration data from flat runs. We anticipate that our extension of these results will enable us to predict locomotion performance on complex dune environments in the field in near real time using only a high-speed camera and data directly available to the robot.

BibTeX entry
	title = {{R}{H}ex for remote desert sensing},
	booktitle = {NRI PI Meeting},
	author = {Roberts, Sonia F. and Koditschek, Daniel E.},
	year = {2015}

Copyright Kodlab, 2017