EgoTactile: Learning Grasp Pressure for Everyday Objects from Egocentric Video

A new benchmark called EgoTactile aims to estimate full-hand grasp pressure from egocentric video, addressing a gap in vision-based tactile sensing that has largely been limited to planar surfaces or fingertip contacts, according to a paper submitted on 8 Jun 2026 [1][2]. The benchmark pairs egocentric video with full-hand pressure supervision for diverse everyday objects and includes a bare-hand transfer subset to support generalization to natural, unconstrained scenarios [1][2]. The researchers note that existing methods often depend on intrusive hardware for dense tactile sensing, which limits their applicability in immersive virtual reality and robotic manipulation [2]. To establish a performance baseline, the team first developed EgoPressureFormer, a discriminative model [2]. They then introduced EgoPressureDiff, a conditional diffusion framework that adapts a large-scale pre-trained video diffusion backbone [2]. The framework incorporates a Physically-Informed Feature Rectification layer designed to inject semantic constraints, helping the system resolve visual-physical ambiguities and infer plausible contact patterns from partial observations [2]. The work appears on arXiv, an open-access repository for electronic preprints that has hosted more than two million articles since its launch in 1991 and currently receives about 24,000 submissions per month [6]. The paper was submitted to the Computer Vision and Pattern Recognition section [1]. In extensive experiments, EgoPressureDiff demonstrated superior performance on the benchmark and showed robust transferability to in-the-wild settings, according to the authors [2]. The project page is available at the EgoTactile website [2]. The release comes as arXiv continues to expand community-driven tools through its arXivLabs framework, which allows third-party collaborators to build features such as bibliographic explorers and code finders that integrate directly with abstract pages [5].