A Bio-Inspired Data-Driven Locomotion Optimization Framework for Adaptive Soft Inchworm Robots

Abstract

This paper presents a data-driven framework for optimizing energy-efficient locomotion in a bio-inspired soft inchworm robot. Leveraging a feedforward neural network, the proposed approach accurately models the nonlinear relationships between actuation parameters (pressure, frequency) and environmental conditions (surface friction). The neural network achieves superior velocity prediction performance, with a coefficient of determination (R2) of 0.9362 and a root mean squared error (RMSE) of 0.3898, surpassing previously reported models, including linear regression, LASSO, decision trees, and random forests. Particle Swarm Optimization (PSO) is integrated to maximize locomotion efficiency by optimizing the velocity-to-pressure ratio and adaptively minimizing input pressure for target velocities across diverse terrains. Experimental results demonstrate that the framework achieves an average 9.88% reduction in required pressure for efficient movement and a 6.45% reduction for stable locomotion, with the neural network enabling robust adaptation to varying surfaces. This dual optimization strategy ensures both energy savings and adaptive performance, advancing the deployment of soft robots in diverse environments.