A Biomimetic Framework for Dynamic Simulation and Heuristic Grasp Planning of Soft Objects in the Presence of External Force

<jats:p>Robotic manipulation of soft, deformable objects is harder than rigid-body grasping: the object's many degrees of freedom make real-time contact prediction expensive, and grasp stability depends on the coupled dynamics of gripper, surface, and external load. This paper presents a unified, biomimetic framework with three components. A three-dimensional mass–spring–damper (MSD) model of the soft object is integrated in state-space form with a fourth-order Runge–Kutta scheme for near-real-time simulation. A five-component heuristic scores candidate grasps from force balance, contact-triangle regularity, centroid proximity, and penalties for sharp vertices and sharp-edge proximity. Normal and tangential contact follow a Hunt–Crossley plus Coulomb stick–slip model whose six parameters are identified by Particle Swarm Optimization (PSO). The framework is validated physically in three stages. A bench-test calibration on four soft materials (19,168 fitted events) confirms the contact exponent (n = 0.46 ± 0.07, force RMSE 2.7%). A 1755-trial campaign on a physical three-finger gripper succeeds on 86.5% of the proposed planner's validation-set grasps (77.8% over all 819 of its trials, including an adversarial set) and predicts the trial-mean fingertip normal force without systematic bias (slope 0.98, R² = 0.72). Finally, a 1230-trial disturbance campaign addresses manipulation under external force: the calibrated friction budget (closed-form statics on the planned contacts) predicts the load a grasp resists before slipping. The planner holds about 86% of disturbances within its natural-friction operating envelope, and 80.8% over all 830 of its trials (1230 across planners), and predicts the lateral slip threshold to a median error of 10.7% — versus 74% for a geometric baseline — with a conservative bias (R² = 0.56, dominated by a minority of large-error trials).</jats:p>