Consistency-based Gait Generation for Terrain-blind Bipedal Walking via V-SLIP Model and Reinforcement Learning*

Terrain-blind walking ability is significant for biped robots to address unforeseen obstacles and terrain modeling errors. To achieve the stable and robustness walking on unknown terrain, this paper proposed a consistency-based gait generation method for terrain-blind bipedal walking via the V-SLIP model and Reinforcement learning method. Firstly, a real-time gait generation scheme based on the Variable Spring-Loaded Pendulum (V-SLIP) model is proposed, which considers a consistency-based gait generation strategy. Secondly, a reinforcement learning-driven controller is designed to regulate the motion trajectory, which is realized by adjusting leg stiffness values to compensate for unexpected terrain height changes. Finally, the proposed method is verified through simulation experiments. Experimental results indicate that the stable terrain-blind walking via V-SLIP model with a random terrain height change of 0.05 m and a walking speed of $1 \mathrm{~m} / \mathrm{s}$ is realized.