Towards Logic-Aware Manipulation: A Knowledge Primitive for VLM-Based Assistants in Smart Manufacturing

TL;DR

This paper introduces a formal manipulation-logic schema {} for vision-language models in manufacturing, enabling better handling of contact-rich actions through object-centric knowledge representation and reasoning.

Contribution

It formalizes an object-centric manipulation-logic schema {} as a knowledge primitive for VLM-based assistants, supporting planning, data augmentation, and retrieval in manufacturing tasks.

Findings

Schema improves VLM planning quality in manufacturing tasks.

Knowledge base enhances data augmentation and retrieval.

Demonstrated on a 3D-printer spool-removal task.

Abstract

Existing pipelines for vision-language models (VLMs) in robotic manipulation prioritize broad semantic generalization from images and language, but typically omit execution-critical parameters required for contact-rich actions in manufacturing cells. We formalize an object-centric manipulation-logic schema, serialized as an eight-field tuple {\tau}, which exposes object, interface, trajectory, tolerance, and force/impedance information as a first-class knowledge signal between human operators, VLM-based assistants, and robot controllers. We instantiate {\tau} and a small knowledge base (KB) on a 3D-printer spool-removal task in a collaborative cell, and analyze {\tau}-conditioned VLM planning using plan-quality metrics adapted from recent VLM/LLM planning benchmarks, while demonstrating how the same schema supports taxonomy-tagged data augmentation at training time and logic-aware retrieval-augmented prompting at test time as a building block for assistant systems in smart manufacturing enterprises.