Time Efficiency Improvement in Quadruped Walking with Supervised Training Joint Model

To generate stable walking of a quadruped, the complexity of the configuration of the robot involves a significant amount of optimization that decreases to its time efficiency. To address this issue, a machine learning method was used to build a simplified control policy using joint models for the supervised training of quadruped robots. This study considered 12 joints for a four-legged robot, and each joint value was determined based on the conventional method of walking simulation and prepossessed, equaling 2508 sets of data. For data training, the multilayer perceptron model was used, and the optimized number of epochs used to train the model was 5000. The trained models were implemented in robot walking simulations, and they improved performance with an average distance error of 0.0719 m and a computational time as low as 91.98 s.

Automated summary

To achieve stable walking for a quadruped robot, optimization is necessary but may decrease its time efficiency. In this study, a machine learning method was used to develop a simplified control policy using joint models for supervised training. The trained models improved walking performance with an average distance error of 0.0719 m and a computational time as low as 91.98 s.