Adaptable and Robust Multi-Robot Decision Making through Generalized Sequential Stochastic Task Assignment

点击翻译按钮获取中文摘要

This project envisions teams of multiple robotic systems cooperatively and autonomously executing complex missions in the physical world. These missions include environmental monitoring, search and rescue, and scientific exploration, where robots are tasked to provide timely information to end users. The success of many robotic missions depends on a small number of critical, irreversible, and high-impact decisions. Examples of such decisions include selecting where to deploy aerial robots from ground robots, determining how to deploy communication hardware, and specifying when to execute specific motion behaviors. In current fielded robotic systems, these types of decisions are largely left to human operators, who typically do not have the required situational awareness, reasoning skills, or available time to make these decisions effectively. This work seeks to bridge the gap between current multi-robot systems that require significant human input to make high-impact decisions and future intelligent robotic systems capable of executing the most effective behaviors at the right time and location.The ultimate objective of this project is to develop new algorithmic solutions for making high-impact decisions in heterogeneous multi-robot teams. When reasoning over such decisions, many variables must be considered, such as the mission goals, available actions, environment belief models, future rewards, and the behaviors and capabilities of other robots. Many of these variables carry a significant degree of uncertainty, have a prior belief of their value, and may change based on dynamic conditions and robot observations. New algorithmic solutions for reasoning over this information will be developed by formulating and solving new generalizations of the sequential stochastic assignment problem (SSAP). These SSAP generalizations require reasoning over the uncertain future values of robot actions, accounting for information acquired in situ, exploiting dependencies in the reward distributions, and computing policies in a decentralized manner. Validation will be performed through both simulated and field experiments for marine monitoring and environment exploration scenarios. The developed algorithms will be made publicly available through open source distribution and will help foster ongoing collaborations with marine and environmental scientists.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.