RI: Small: Collaborative Research: Information-driven Autonomous Exploration in Uncertain Underwater Environments

Recent rapid advances in unmanned ground, aerial, and underwater vehicles promise a myriad of exciting applications in environmental monitoring, search and rescue, and mobility for goods and humans. A core requirement of all these applications is the capability of autonomously navigating through uncertain environments, to maximize information gain while accommodating motion, time, and energy constraints. The goal of this proposal is to develop a theoretical and algorithmic framework for computationally efficient control synthesis that enables information-driven exploration of environments for autonomous robots. The proposed approach is centered around the concept of ergodic exploration, such that the time a robot or robots spend at a particular location is proportional to the expected information density at that location. Sequential action control will be used as the optimization engine for ergodic exploration to achieve real-time control synthesis. Novel analytical tools will be explored to understand and establish the stability and convergence properties of the proposed algorithms under dynamic, evolving environments. Distributed algorithms will be developed to allow individual robots to synthesize their own controllers while sharing essential centralized information. For sensing applications involving uncertain flow environments, ergodic exploration with active probing is proposed. The proposed theory and algorithms will be experimentally evaluated in lakes, where a group of gliding robotic fish will be used to map harmful algal blooms and localize the source of chemical spills. Keywords: active sensing; underwater robotics; autonomous exploration; nonlinear control; collaborative sensing. The proposed work will advance recently developed techniques in ergodic control, where trajectories are based on their spatial statistics instead of their time evolutions. This enables the automated design of trajectories that sample throughout high information density regions of the state, and extends to general nonlinear systems. In particular, this work will enable fundamental understanding of the stability of the combined physical and information states to provide correct-by-construction real-time control for ergodic exploration. The project will result in a set of field-tested algorithms that guarantee effective information gathering in uncertain environments. The proposed work is of immediate relevance to a number of applications of societal impact ? it will facilitate the effective use of autonomous robots in environmental monitoring, search and rescue, surveillance and security, among others. The project will provide exciting training opportunities for graduate and undergrad students, especially those from underrepresented groups, and enrich course offerings at both Northwestern University and Michigan State University. The proposed research will be showcased in the Museum of Science and Industry in Chicago, during a National Robotics Week exhibit in the main rotunda of the museum with an estimated viewership of over ten thousand on-site visitors. It will also result in an open-source robotic fish education kit, along with design competitions, to energize K-12 students to pursue science and engineering. Aside from publications and conference presentations, the outcomes of this research will be directly transferrable to PI Murphey?s other projects with the Rehabilitation Institute of Chicago and the Northwestern medical school. The project will significantly advance gliding robotic fish?s autonomy in mobile sensing and thus