Abstract
The paper presents a novel design for a soft bio-mimetic finger and soft thumb structure for bionic hand applications. It introduces an anthropomorphic pneumatic flexible finger system using a PneuNets framework to enhance flexibility and maneuverability. The research investigates the influence of geometric variations (wall thickness, chamber number, and spacing) on finger deformation, demonstrating that reduced wall thickness and augmented chambers substantially improve flexibility. A key innovation is the soft thumb design that accurately replicates the complex movements of the Carpometacarpal (CMC) joint, enabling natural opposition and dexterity. Eight models were developed for four fingers and two models for the thumb. Simulation results indicate that models with thinner walls (2 mm) achieve bending angles exceeding 80° at 120 KPa, whereas 3 mm models remain below 50°. Moreover, increasing the number of chambers enhances deformation, with each added chamber contributing approximately 41 % more flexibility. For the thumb models, we successfully mapped the motion ranges and accurately mimicked the base joint, enabling natural opposition and dexterity. Furthermore, the paper also integrates Artificial Neural Networks (ANNs) to model forward kinematics, improving the estimation of bending angles and end-tip positions, which enhances the overall adaptability and control of the system.