GraspFoM: Towards Reconstruction-Driven Robotic Grasping with 3D Foundation Priors

A new framework called GraspFoM unifies 3D object reconstruction and robotic grasp prediction using shared foundation-model priors, achieving state-of-the-art results with only a small number of additional trainable parameters, according to a paper submitted to arXiv on June 7, 2026 [1][2]. The work addresses a persistent challenge in robotic manipulation: grasping objects under partial observation. Reliable grasping depends on both local contact cues and object-level 3D structure, yet existing geometry-aware methods typically treat reconstruction as an intermediate prediction rather than a reusable prior for grasping [2]. GraspFoM instead leverages 3D foundation priors from SAM3D to build a shared 3D object latent that serves both reconstruction and grasp pose prediction [1][2]. Built on this shared latent, the framework introduces an anchor-initialized truncated pose-reasoning diffuser that predicts continuous and multimodal grasp poses without relying on discrete grasp candidates [2]. The interaction between the two tasks is further refined through a reconstruction-aware scorer and a residual latent updater: reconstruction provides grounded geometric cues, while grasp supervision refines the shared object latent toward grasp-relevant affordances [2]. The system jointly predicts grasp poses and reconstructs high-fidelity 3D assets in both mesh and 3D Gaussian Splatting forms [2]. Comprehensive experiments show state-of-the-art performance on both reconstruction and grasping benchmarks, with component-wise ablation studies confirming the contribution of each module [2]. The paper was posted on arXiv, the open-access e-print repository that hosts preprints across physics, computer science, and related fields and receives approximately 24,000 submissions per month as of late 2024 [6]. The abstract page for the paper includes integrations from arXivLabs, a framework launched in 2020 that allows community collaborators to develop experimental tools such as bibliographic explorers and code finders directly on the site [5][4].