NRI: FND: A Framework for Human-Team-Supervised Autonomy with Application to Underwater Search and Rescue

点击翻译按钮获取中文摘要

Advances in computing and manufacturing have led to rapid developments in autonomous robots. For sophisticated tasks such as search and rescue, it is often critical to integrate human knowledge and perception skills with the capabilities offered by robots. Taking underwater search and rescue as a motivating context, this project focuses on developing a principled design framework for optimizing the performance of a mixed human-robot team comprised of multiple human operators and heterogeneous robots. By enabling efficient and reliable human-robot interactions, this work will facilitate the use of robots in hazard response, environmental monitoring, mobility of goods and humans, healthcare, manufacturing, and many other applications of societal impact. The project will provide training opportunities for graduate and undergrad students, including those from underrepresented groups. It will also provide research training to high school students and K-12 teachers. An open-source robotic fish educational kit and demos of EEG-mediated human-robot interactions will be developed to pique the interest of K-12 students in science and engineering. The project will further produce an underwater robotics testbed available for use by the broader robotics and control community.This research will develop a generalizable framework for rigorous and systematic design of autonomy supervised by a team of interacting human operators, which will enable the leveraging of human operators' adaptivity in complex scenarios while mitigating performance deterioration due to loss of situational awareness. The framework will consist of two tightly coupled modules. The first module will involve optimal task allocation and scheduling for event-triggered human team supervision, which will be formulated as a semi-Markov decision process (SMDP) for a complex queueing network capturing task processing by a team of human operators with different skill sets. Human cognitive dynamics will be incorporated via practical models, and efficient algorithms for solving the SMDP are examined while uncertainties introduced by stochasticity in cognitive processes and variability among human operators are accommodated. The second module of the framework will deal with informative path planning for autonomous robots that optimally balances the explore-exploit trade-off in their search for targets of interest, by solving a multi-armed bandit problem that incorporates mobility constraints of the robots. The framework will be experimentally evaluated in field trials emulating underwater search and rescue, which will involve a group of gliding robotic fish and remotely operated vehicles (ROVs), supervised by a team of two human operators.