CAREER: Autonomous Live Sketching of Dynamic Environments by Exploiting Spatiotemporal Variations

点击翻译按钮获取中文摘要

There are many emergency scenarios that require humans to understand the environments instantly. For example, after disasters of chemical or nuclear leakage, the spread and intensity of the contaminant need to be characterized immediately for the best first response. Inspired by artistic live sketching which needs to rapidly capture transient scenes and unveil their most salient spatiotemporal characteristics, the autonomous mobile robots equipped with advanced artificial intelligence (AI) algorithms will be utilized to "live sketch" highly dynamic environments through autonomous contaminant sampling and real-time environmental modeling. Success of this research could potentially be a game-changer for automated environmental monitoring under extreme conditions. The results can be naturally extended to many applications including those non-disastrous but time-critical scenarios such as smog pollution and algal bloom monitoring. The education objective is to promote general interest in robotics careers by integrating this research into curriculum development, direct student involvement in research (particularly for female and minority students), as well as community outreach.The research objective of this project is to investigate principled, expeditious, and precise environmental modeling techniques through adaptive environmental sampling with robotic vehicles. This research program tackles a variety of needed techniques drawn from important AI-robotics subfields including data-driven modeling, sampling trajectory planning, decision making under uncertainty, as well as multi-robot coordination. An important goal is to obtain deeper insights into all these subfields and their connections, leading to a design of a principled and comprehensive framework for building a complete integrated system. New solutions of a set of inter-dependent modeling and optimization methods will be developed, so that the latent environment model and its spatiotemporal variations (e.g., contamination distribution and diffusion processes) can be learned with high accuracy, even through using a small number of samples constrained by a very short data-collection time window. The proposed efforts include development of theoretical results and algorithms, but also emphasize their applications in various challenging and unstructured environments.This project is supported by the cross-directorate Foundational Research in Robotics program, jointly managed and funded by the Directorates for Engineering (ENG) and Computer and Information Science and Engineering (CISE).This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.