Related Experiment Video
Updated: May 5, 2026
06:17
Deep-Learning Based Multi-Joint Synchronous Tracking for Objective Quantification of Hindlimb Locomotor Kinematics in Rats
Published on: April 3, 2026
116
Gait Optimization Control of Spinal Quadruped Robot Based on Deep Reinforcement Learning
Guozheng Song1,2, Qinglin Ai1,2, Lin Li1,2
1College of Mechanical Engineering, Zhejiang University of Technology, Hangzhou 310023, China.
Sensors (Basel, Switzerland)
|May 4, 2026
Summary
This study introduces a spinal joint for quadruped robots, enhancing flexibility. A deep reinforcement learning framework combining central pattern generators (CPG) and TD3 algorithms optimizes gait for improved stability and terrain adaptability.
Area of Science:
- Robotics
- Bio-inspired engineering
- Artificial Intelligence
Background:
- Quadrupedal locomotion relies on spinal flexibility for natural movement and adaptability.
- Existing quadruped robots often lack the dynamic range of motion seen in biological systems.
- Integrating spinal actuation offers potential for enhanced mobility and posture control in robotic platforms.
Purpose of the Study:
- To develop and validate a novel control strategy for a quadruped robot with an actuated spinal joint.
- To improve the gait stability and terrain adaptability of spinal quadruped robots in complex environments.
- To explore the integration of central pattern generators (CPG) with deep reinforcement learning for optimized robotic locomotion.
Main Methods:
- Designed a quadruped robot model incorporating an actuated spinal joint and analyzed its parameters.
- Developed a central pattern generator (CPG) coupling model that integrates spinal motion parameters.
- Implemented a deep reinforcement learning framework combining CPG with the twin delayed deterministic policy gradient (TD3) algorithm, utilizing a joint incremental strategy for gait optimization.
- Conducted simulations and experiments on various obstacle terrains to evaluate the proposed TD3-CPG algorithm.
Main Results:
- The proposed TD3-CPG algorithm effectively optimized the gait of the spinal quadruped robot.
- Significant improvements were observed in walking stability, speed, and terrain adaptability across diverse obstacle terrains.
- The integration of the spinal joint and the advanced control framework demonstrated enhanced dynamic motion and posture adjustment capabilities.
- Experimental validation confirmed the algorithm's effectiveness in real-world scenarios.
Conclusions:
- The developed deep reinforcement learning framework (TD3-CPG) successfully enhances the locomotion capabilities of spinal quadruped robots.
- The bio-inspired spinal joint, coupled with intelligent control, significantly improves robot performance in challenging environments.
- This research provides a promising approach for developing more agile and adaptable legged robots for various applications.
