Phy-Tac: Toward Human-Like Grasping via Physics-Conditioned Tactile Goals

TL;DR

This work addresses the gap in robotic grasping where force regulation is typically reactive and excessive. It introduces Phy-Tac, a unified framework that couples physics-informed pose planning, tactile imprint prediction via a physics-conditioned latent diffusion model (Phy-LDM), and a latent-space LQR controller to realize force-optimal stable grasping (FOSG). A physics-conditioned tactile dataset supports learning, and Phy-LDM generates tactile goals that guide proactive, minimal-contact-force manipulation. Empirical results across rigid and compliant objects show improved grasp stability and force efficiency over baselines, highlighting a step toward human-like tactile intelligence in robotic hands.

Abstract

Humans naturally grasp objects with minimal level required force for stability, whereas robots often rely on rigid, over-squeezing control. To narrow this gap, we propose a human-inspired physics-conditioned tactile method (Phy-Tac) for force-optimal stable grasping (FOSG) that unifies pose selection, tactile prediction, and force regulation. A physics-based pose selector first identifies feasible contact regions with optimal force distribution based on surface geometry. Then, a physics-conditioned latent diffusion model (Phy-LDM) predicts the tactile imprint under FOSG target. Last, a latent-space LQR controller drives the gripper toward this tactile imprint with minimal actuation, preventing unnecessary compression. Trained on a physics-conditioned tactile dataset covering diverse objects and contact conditions, the proposed Phy-LDM achieves superior tactile prediction accuracy, while the Phy-Tac outperforms fixed-force and GraspNet-based baselines in grasp stability and force efficiency. Experiments on classical robotic platforms demonstrate force-efficient and adaptive manipulation that bridges the gap between robotic and human grasping.

Figures (11)

• Figure 1: Comparison of human grasping strategy (a) and our Phy-Tac framework (b) for FOSG. Humans first select an optimal contact pose (S1), estimate the required grasping force (S2), and refine it to a just-enough level to maintain stability (S3). Likewise, Phy-Tac unifies pose planning (S1), tactile state prediction (S2), and force regulation (S3) to achieve the same principle, where S2 and S3 form a closed loop for optimal force regulation.
• Figure 2: The description of human-inspired stable grasping method Phy-Tac. a). Our physics-conditioned SG dataset consists of objects' $3$D texture models, contact region information, control signals, and tactile states. b). Our force-optimal stable grasping strategy Phy-Tac contains 3 steps, i.e., grasping pose planning, grasping state estimation, and contact force regulation.
• Figure 3: A example to show the influence of physical condition for contact tactile state. Specifically, the physical factors are mass $M$, texture $T$, and contact region $P_t$.
• Figure 4: The contact description of the selected candidates with top score. The first row in each sub-figure is the grasping pose generated by GraspNet for bread. The second row is the depth information of contact region in left/right finger for generated grasping pose.
• Figure 5: The parameter value of the selected grasping candidates generated by GraspNet. a). The score and evaluating metrics of each candidates for grasping pose planning. b). The value of mismatching rate is used to select suitable candidate numbers in contact-optimal pose planning process.