CAREER: Interactive Decision-making and Resilient Planning for Safe Legged Locomotion and Navigation.

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

This Faculty Early Career Development (CAREER) project aims to bridge fundamental gaps between state-of-the-art robot task and motion planning and the ultimate goals of safe and autonomous robot locomotion. This work will resolve major computational hurdles that have hindered the use of symbolic planning and decision-making methodologies on the highly nonlinear, hybrid, and under-actuated legged systems, reducing their ability to reliably operate in unstructured environments, efficiently collaborate with robot teammates, and safely interact with humans. Specific long-standing challenges to be addressed are robustness to contact terrain uncertainty, locomotion task generalization, navigation safety coupled with complex locomotion dynamics, and proactive interaction with pedestrians to avoid collisions. This project will move legged robotic systems from the research laboratory, and into challenging application domains such as disaster first responders, surveillance in civil and mechanical infrastructures, and planting in agricultural environments. The research is complemented by education activities to (i) integrate undergraduate students into a new experiment-focused research program through the multi-semester Vertically Integrated Program (VIP) at Georgia Institute of Technology; (ii) host high school students from Atlanta minority-serving public schools to perform research and enhance underrepresented population participation in the greater Atlanta area; and (iii) broaden the societal impact by disseminating the integrated robotics research through an interactive workshop. This project will develop a novel task and motion planning framework for bipedal locomotion interacting with complex environments. In particular, it will retain and augment the strengths of full-body-dynamics-aware trajectory optimization (TO) with the complementary capabilities of symbolic planning and policy learning for environment abstraction, formal task specification design, and robust decision-making. This project has a large focus on experimental evaluations to enable transformative new legged navigation functionalities in real-world environments. The research outcome will enable (i) safe navigation in partially observable environments and proactive interaction with pedestrians while incorporating low-level locomotion dynamics constraints, (ii) versatile contact behavior planning and robustness reasoning through risk-sensitive TO and adversarially regularized policy learning, and (iii) real-time locomotion failure recovery capability via behavior trees to address unexpected environment interventions. This project will advance the state of the art in safe and autonomous navigation of legged systems and pave the road for future studies of heterogeneous robot teaming. 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.