SYNLOCO: Synthesizing Central Pattern Generator and Reinforcement Learning for Quadruped Locomotion

Abstract

The Central Pattern Generator (CPG) is adept at generating rhythmic gait patterns characterized by consistent timing and adequate foot clearance. Yet, its open-loop configuration often fails to adjust the system’s control performance in response to environmental variations. On the other hand, Reinforcement Learning (RL), celebrated for its model-free properties, has gained significant traction in robotics due to its inherent adaptability and robustness. However, initiating traditional RL approaches from the ground up presents a risk of converging to suboptimal local minima and slow learning convergence. In this paper, we propose a quadruped locomotion framework-called SYNLOCO-by synthesizing CPG and RL, which can ingeniously integrate the strengths of both methods, enabling the development of a locomotion controller that is both stable and natural with partial state observations (e.g., no velocity measurements). To optimize the learning trajectory of SYNLOCO, a two-phase training strategy is presented. Both ablation analysis and experimental comparison are performed using a real quadruped robot under varied conditions, including distinct velocities, terrains, and payload capacities. The experiments showcase SYNLOCO’s efficiency in producing consistent and clear-footed gaits across diverse scenarios, despite no velocity measurements. The developed controller exhibits resilience against substantial parameter variations, underscoring its potential for robust real-world applications.