SAM-ACE: Situation-aware Hierarchical Hybrid LeArning-based Modeling and Robust Adaptive Control of Autonomous LEgged Robots

Existing quadruped robots have limited adaptability for challenging environments (such as highly uneven, obstructed, and uncertain terrains) and suffer from limited agility due to unaffordable computational complexity for onboard systems and slow mobility caused by inefficient real-time sensing capability, which hinders their applications in military operations. To address these weaknesses, it is critical to developing novel techniques that can (1) effectively model the real-time quadruped robot dynamics and precisely represent the impacts of uncertain terrains and various tasks, (2) efficiently control quadruped robots across challenging terrains that have been impassable by the existing autonomous agents, and (3) truly integrate emerging data-driven learning techniques into real-time model-based control to harvest benefits from both sides. Built on the advanced sensor fusion techniques developed by Intelligent Fusion Technology, Inc. (IFT) and optimization and control theory by the University of Nevada, Reno (UNR), IFT with UNR proposes to develop a new-generation intelligent learning-based model predictive control algorithm, including situation-aware hierarchical bi-level hybrid dynamic modeling, hierarchical hybrid robust adaptive control, and physics-guided and low-computational-cost hybrid learning, to boost the existing legged robotics technology and ensure U.S. leadership in the legged robotics system development. The expected outcomes of this effort include a novel and solid theoretical foundation to address the robust and optimal control problem for autonomous legged robots during discontinuous events and efficient hierarchical hybrid modeling and control algorithms that can advance legged robotics system development. ???????