Learning-Aided Integrated Control and Semantic Perception Architecture for Legged Robot Locomotion and Navigation in the Wild

点击翻译按钮获取中文摘要

This project develops an open-source system to allow legged robots to perform autonomous exploration and environmental monitoring in unstructured environments, such as in a forest. A central theme is the design and deployment of a unified and integrated framework that exploits the redundancy of onboard sensory modalities and cross-integrates with feedback control and planning. The project seeks major advances in control theory, computer vision, embedded systems, and planning under uncertainty, areas that are often studied in isolation. By studying these areas as a systems problem, the researchers expect significant advances for autonomous systems in unstructured environments. The results of the research are demonstrated on a Digit-series 3D biped and a Mini Cheetah quadruped robot. The research elevates the state of the art in deploying autonomous mobile robots “in the wild.” In off-road and unstructured settings, the main technique currently employed by the autonomous vehicle industry, registering into known high-definition maps on the basis of collected sensor measurements, is not possible and adversely affects autonomy. The project develops a real-time multi-layer dense semantic occupancy mapping with an extended set of terrain labels and a “walk-ability” index combined with an integrated motion planner for an autonomous system. The traversability map enables a walking robot to make dynamic real-time planning and feedback control decisions, adjusting for gait characteristics and for precise foot placement based on the surrounding terrain and environment.The principal investigators are co-developing a freshman college course to inspire students by teaching how practicing engineers employ computational linear algebra to solve large and important problems as arise in the study of autonomous robots. Course projects are selected from contemporary topics in robotics such as, for example, map building from LiDAR point clouds, machine learning for spatial representation of data, and feedback control of mobile platforms. In addition, the project engages in outreach to underrepresented communities through collaborative educational instruction in STEM fields in Detroit public high schools.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.