Online Hierarchical Planning for Multicontact Locomotion Control of Quadruped Robots

Owing to challenges such as lengthy solving time and convergence issues, multicontact locomotion planning problems are often formulated with fixed contact schedules, greatly restricting the flexibility of quadrupedal robot behavior. This article presents a novel hierarchical planning framework designed for online multicontact locomotion control of quadruped robots. At the top level, we systematically explore the gait branches of a passive planar quadrupedal dynamic model via numerical continuation. In addition, we propose a gait assessment strategy at varying speeds, comprehensively considering both energy consumption and stability. At the middle level, we present an efficient strategy for contact-implicit optimization problems by integrating McCormick envelopes and alternating direction method of multipliers. Using the gait selection reference obtained from the top level as an initial guess can substantially reduce the solution space and bring the resulting solution closer to the global optimum. Based on the gait pattern and state trajectory reference acquired from the middle level, we adopt a hybrid kinodynamic model for application in model predictive control of quadrupedal locomotion. To validate the proposed hierarchical planning framework, we conduct comparative locomotion experiments on the quadrupedal robot SCIT-Dog under varying speeds. Experimental results demonstrate the effectiveness and superiority of the proposed algorithm compared to the impulse-based gait transition method and the predefined trot gait pattern. Moreover, the observed gaits align with those of quadrupedal animals, demonstrating the potential of the proposed framework to enhance adaptability and performance in multicontact locomotion planning for quadrupedal robots.