Robots in the Wild: Collaborative Exploration and Mapping
Robotics and AI are enabling safe and efficient exploration of unstructured, uncertain, and hazardous environments, driving new discoveries in earth and space sciences alike. In this talk, I will present examples from three domains -- marine sciences, agriculture, and geology, highlighting spatio-temporal observation of environmental and biological processes at unprecedented scale and resolution. First, I will describe a methodology for optimal water sample collection by autonomous underwater vehicles for marine ecosystem monitoring. Principled approaches from Bayesian optimization and optimal stopping theory facilitates improved sample quality across subsequent missions. Next, I will describe how we are using drones and specialized sensor suites for fruit counting and crop health monitoring, enabling growers and agronomists to work towards improved yield and reduced water and herbicide usage. I will close with recent results in geomorphological analysis using structure from motion and deep learning on close-range aerial imagery. Rocks annotated by human expert on a small dataset of images are used to train deep neural networks that automatically extract semantic information (rock boundaries) on large volumes of unlabeled high-resolution aerial imagery. Further post-processing using structural analysis and shape descriptors results in estimates of geometric traits, such as rock diameter and orientation. The spatial distribution of rock traits may offer new insights into geological processes, and we foresee our pipeline being applicable to a variety of other tasks such as forest mapping, and damage assessment after natural disasters.
Speaker Bio:
Jnaneshwar Das holds the Alberto Behar Research Professorship at the School of Earth and Space Exploration, with expertise in robotic monitoring, machine learning, autonomous systems, and unpiloted vehicles. His research contributions include mixed-initiative spatio-temporal observation of environmental and biological processes, exploiting mathematical models, and algorithms for closing the loop on data-driven robotic sampling.
Prior to joining ASU, Das was a postdoctoral researcher at the General, Robotics, Automation, Sensing and Perception (GRASP) laboratory at the University of Pennsylvania, where he investigated the use of unmanned aerial vehicles (UAVs) for precision agriculture, Earth sciences, and humanitarian applications.
He is also a co-organizer of the National Science Foundation’s Student Cyber-Physical Systems (CPS) Challenge, and the founder of the OpenUAV Project, producing testbeds to support UAV education and research.