CAREER: Human-Inspired Multi-Robot Navigation

点击翻译按钮获取中文摘要

Indoor mobile robots are increasingly becoming a part of our lives. Whether there are Roombas cleaning the floor or Kiva robots delivering parts in warehouses, the robots should be able to avoid collisions while successfully completing their tasks. However, despite the maturity of existing motion planning techniques and the recent rise of learning and big data techniques, mobile robots still lack the decision making ability of humans. This Faculty Early Career Development (CAREER) project will develop techniques for efficient and socially intelligent multi-robot navigation, shaping the next generation of mobile robots that can reason about how their actions influence the other agents present in the scene and act accordingly, much like humans do. The resulting advances will facilitate the successful deployment of "thinking" mobile robots that can be seamlessly integrated into our homes and workspaces. This research spans across different areas, including motion planning, machine learning, and reinforcement learning. With its interdisciplinary nature and relevance for modern technologies, it is ideal for inspiring the next generation of students and exposing the broader community to STEM areas couched in progressive applications in robotics and AI. The project includes integrated educational, research, and outreach activities for K-12, undergraduate, and graduate students, promoting a high level of participation by women and underrepresented minorities, and developing new courses and updated curricula related to robotics.This project will introduce a human-inspired paradigm shift in the design of multi-robot navigation algorithms. Humans know when they have to be polite and yield to others and when to take decisive actions, efficiently performing complex navigation tasks without collisions. The objective of this project is to enable such behavior on mobile robots by leveraging publicly available human-human interaction data and our own human-robot interaction experiments along with coupling motion planning with learning techniques. Specifically, the project will focus on two two inter-related research thrusts that will lead to i) new algorithms that take advantage of human trajectory datasets to learn what controls humans take in different interaction scenarios; ii) new approaches that enhance existing local navigation planners with the learned controls to enable human-like decision making; iii) a reinforcement learning framework for multi-robot navigation that generalizes robot navigation policies to unknown interactions scenarios; iv) new datasets involving interactions between humans and robots, and subsequently v) new algorithms for multi-robot navigation in human-populated environments. This work will be evaluated both in simulation and on real robots, and related algorithms and datasets will be made publicly available to facilitate further research and exploration by the robotics and AI community. If successful, this project will shape the next generation of indoor mobile robots that can enrich our quality of life and work, and has the potential to significantly benefit society through its integrated education plan.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 project is jointly funded by CISE/IIS, the Established Program to Stimulate Competitive Research (EPSCoR), and ENG/CMMI.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.