Coordinating Spinal and Limb Dynamics for Enhanced Sprawling Robot Mobility

Abstract

Sprawling locomotion in vertebrates demonstrates how spinal mobility improves stability, maneuverability, and adaptability across complex terrain. This work introduces a hybrid locomotion framework that combines biologically inspired Hildebrand gait generation with deep reinforcement learning to control a salamander-inspired quadruped robot. Extensive simulation and real-world experiments show that integrating structured gait design with learning-based spinal control improves robustness, locomotion efficiency, and sim-to-real transfer compared with purely learning-based approaches.