Slip‐Adaptive Neural Control of Gecko‐Inspired Adhesive Robots

Gecko‐inspired robots outperform other climbing robots with their high‐efficiency dry‐adhesion mechanism and superior terrain adaptability. However, a lack of precise adhesive‐force sensing and control compromises their locomotion stability under severe disturbances such as slipping. This article presents a neural adhesion controller that enables a gecko‐inspired robot to adapt effectively to slip disturbances. To overcome the challenges associated with sensing and predicting highly nonlinear ground adhesive forces, the controller integrates an echo state network (ESN) with a multilayer perceptron (MLP). It accurately estimates real‐time adhesion in the normal and shear directions via proprioceptive joint torque feedback, achieving a low prediction error (mean squared error (MSE) = 0.06 across limbs). Additionally, the system can forecast incipient slip over consecutive strides by leveraging learned temporal patterns. The proposed controller successfully recovers the climbing robot from destabilizing slip events on a low‐adhesion surface, representing a significant advantage toward robust and reliable real‐world locomotion.