Generalizable pre-grasp force prediction

Establish generalizable pre-grasp force prediction that accurately estimates the minimum feasible grasping force required to lift an object without slip from pre-contact observations across diverse, previously unseen objects and gripper embodiments, including compliant grippers.

Background

The paper highlights that existing analytical methods for grasp force estimation depend on precise prior knowledge of object geometry, friction, and contact conditions, which are rarely available for novel objects. Data-driven policies similarly struggle to generalize due to the diversity of object properties, interaction conditions, and gripper embodiments.

This challenge is particularly acute for compliant grippers, where nonlinear, embodiment-dependent contact mechanics are difficult to model. Consequently, achieving a broadly applicable, pre-contact force prediction strategy that generalizes to unseen scenarios remains unresolved.