FURI | Spring 2026

Learning Spine-Assisted Locomotion: A Reinforcement Learning Comparison of Spined and Rigid Quadruped Robots

Flexible spinal motion improves locomotion efficiency in animals, yet most quadruped robots rely on rigid bodies due to challenges in modeling spine actuation. This project investigates whether spinal flexibility improves reinforcement learning performance in a simulated quadruped robot. A spined quadruped and a rigid-body quadruped are trained under identical conditions, including terrain, reward structure, and control framework. Because tendon-driven actuation is difficult to simulate in IsaacSim, the spine is modeled using a force-based approximation. Performance is evaluated using reward, speed, torque usage, and energy efficiency metrics.