CLARA: A Cognitively-Driven Clarification Framework for Grounding Ambiguous Requests

This research presents CLARA (Conversational cLArification for Robotic Arrangement), a framework designed to handle ambiguous high-level instructions in robotics and AI systems. By proactively eliciting key details from users through targeted questioning, CLARA refines vague directives into structured specifications. These refined instructions guide a diffusion-based image generation pipeline to produce realistic visual renderings of the desired scene, serving as confirmation of the system's understanding. Experimental results and user studies show that CLARA achieves more accurate and consistent outcomes compared to other State-of-the-Art Large Vision-Language Model (LVLM) systems, demonstrating the value of interactive clarification in bridging the gap between vague commands and grounded task execution.