Event

CRAFT-E: An Explainable and Editable Neuro-Symbolic Framework for Affordance-Based Object Retrieval

1 reports · 1 sources

Zhou Chen and other authors propose and deploy the CRAFT-E neuro-symbolic framework for affordance-based object retrieval: it combines a verb-attribute-object knowledge graph derived from large language models, CLIP visual alignment, and energy-based grasp reasoning, and supports targeted edits to the graph directly without retraining the model. The authors also construct a benchmark dataset that uniformly annotates verb-object compatibility, segmentation, and grasp candidates, and deploy the complete pipeline on a physical robot. The authors report that in static scenes, real-world trials involving 20 verbs and 39 objects, ImageNet-based functional retrieval, and COCO-Tasks transfer, CRAFT-E provides a competitive training-free prior for affordance grounding retrieval and can give component-level diagnostics and graph-level interventions. These results are the authors' own claims and have not yet been independently verified.