Improving Decision Making Efficiency in Field Data Collection using a Heterogeneous Group of Legged Robots

Traditional field tests of sediment transport often include construction of wind tunnels to obtain soil erodibility, and collection of large quantities of samples for laboratory analysis of soil properties. These methods are incapable of providing high spatial resolution soil erodibility data, and the delay in obtaining correlations between soil erodibility and environmental control parameters can lead to inefficient decision making in the field.

Recent advancements in legged robots have demonstrated capabilities in providing transformative geoscience field data in-situ with high spatial resolution. In this presentation, we show that a heterogeneous group of collaborative legged robots can also provide different levels of feedback to facilitate field decision making.

We develop two different mobile platforms to assist geoscientists in the investigation of sediment erosion processes: Minitaur, a quadrupedal "scouting" robot, performs fast, dynamic locomotion while using its direct-drive legs as force sensors to assess soil strength, providing a preliminary map of soil erodibility across large areas; RHex, a hexapedal "measurement" robot, performs stable locomotion while carrying multiple scientific sensors and instruments, providing comprehensive assessment of controlling variables such as soil moisture and grain size that correlate with erodibility measurements.

The different types of data provided by the heterogeneous robot group could allow geoscientists to refine their hypotheses and adjust experiment plans in-situ. Initial observations suggest that this can improve the efficiency of field decision making in two ways: (1) uncertainty in the alignment of prior knowledge and the current environment can be reduced through identification of environmental gradients from the scouting robot, allowing for improved hypothesis generation; and (2) hypotheses and experiment plans can be adjusted and refined based on incoming measurements provided by the measurement robot, potentially reducing vulnerability to common human decision biases and helping to improve future predictions.

The presentation will review our progress in developing tests for both hypotheses.