Abstract
Abstract
Soft triaxial tactile sensors typically achieve force decomposition through multi-unit arrays that rely on mechanical priors or through dedicated single-module designs, which must balance structural simplicity and accurate reconstruction across a broad range of loading directions. This paper presents a monolithic laser-induced graphene (LIG) soft triaxial tactile sensor with sequential four-electrode excitation, encoding triaxial contact forces in electrical boundary responses without structural force-decoupling mechanisms. A hysteresis-aware latent-dynamics reconstruction framework explicitly models path-dependent transduction, resolving forces that produce the same electrical response but differ in loading history. A robot-automated calibration and characterization platform with a six-axis force/torque sensor and force-guided loading provides reference triaxial force labels across a densely sampled directional space. The characterization quantifies directional, loading-path, and rate-dependent electrical responses. With pooled calibration at 0.5, 2, and 4~mm/s, the sensor achieves a mean absolute force magnitude error of 0.0528~N on held-out loading directions, with a mean direction error of 5.22$^\circ$ over reference force magnitudes from 0.3~N to 3~N. Baseline comparison on three individually calibrated sensors and ablation studies confirm the advantage of the proposed framework over static and sequential alternatives and identify the latent dynamics accumulator as the primary source of the gain over static reconstruction. The monolithic structure and direct triaxial force output make the sensor suitable for integration at robotic end-effectors and compliant contact interfaces.