Exciting news! We're transitioning to the Statewide California Earthquake Center. Our new website is under construction, but we'll continue using this website for SCEC business in the meantime. We're also archiving the Southern Center site to preserve its rich history. A new and improved platform is coming soon!

Advancing earthquake science with robotics and AI

Jnaneshwar Das, Ramon Arrowsmith, Zhiang Chen, Tyler R. Scott, Chelsea P. Scott, & Devin Keating

Published August 4, 2020, SCEC Contribution #10253, 2020 SCEC Annual Meeting Talk on TBD

Unprecedented progress in AI in recent years has improved data-driven discovery in the earthquake sciences. Together, we envision robotics and AI to advance earthquake science through improved automation and reduced cognitive burden on scientists, while potentially improving the quality of scientific insights drawn from data. Significant work has already been done on AI aided time-series analyses of seismic signals. In our initial SCEC-supported work, application of structure from motion (SfM) and deep learning (DL) for 2D segmentation of rocks has enhanced our understanding of the geomorphology of fault scarps and fragile geologic features including precariously balanced rocks (PBRs), making possible estimation of rock trait distributions over large spatial scales. Traits include size, orientation, and eccentricity. However, constraints in data collection as well as analysis can impact the validity of scientific insights. In addition to AI, robotics tools are needed to mitigate these constraints. First, heteroscedasticity exists due to inconsistent ground sampling resolution when imaging with unpiloted aircraft systems (UAS), propagating into estimated orthorectified maps. Then, estimation of rock traits in 2D, although computationally faster, produces biases due to rock shapes projected to the horizontal plane. This is being addressed through extraction and analysis of 3D traits such as sphericity, accelerated through our existing UAS-SfM-DL pipeline. UAS with close-range imaging and terrain following capabilities are being developed to improve image consistency and to also lead towards mapping of PBRs. Integration of modern automotive LiDARs will enhance map quality, and additionally improve robustness of autonomous navigation algorithms needed for active mapping near rock pillars and PBRs, not to mention applications in rapid post earthquake response mode for the natural and built environment. Finally, inspired by the hazards community that has already started looking at physics engines to study particle behavior under various forcings, we are investigating manipulation of mapped sites and interpretation of surface processes especially particle motions and interactions through physics-based simulations using gaming industry-developed dynamics engines. As a byproduct, science communication and education can be made more engaging leveraging robotic mapping experiments at geologic sites, delivered through end to end simulation testbeds on the cloud

Key Words
earthquake science, robotics, AI

Citation
Das, J., Arrowsmith, R., Chen, Z., Scott, T. R., Scott, C. P., & Keating, D. (2020, 08). Advancing earthquake science with robotics and AI. Oral Presentation at 2020 SCEC Annual Meeting.


Related Projects & Working Groups
Computational Science (CS)