Pipeline
Overview of our proposed framework. Our method for space-aware furniture recommendation and placement consists of scene parsing, rule engine-based prediction, and finally, autoregressive transformer modeling steps.
This paper proposes an agentic artificial intelligence (AI) driven framework for automated furniture recommendation and placement in indoor settings. Unlike existing approaches that treat object selection and spatial arrangement as separate problems, our method unifies space-aware and object-aware synthesis through a two-stage cluster-oriented pipeline. Given a room's floor plan, the first stage uses a proximal policy optimization (PPO) agent integrated with a rule-penalty engine to identify optimal placement regions (``sweetspots'') for furniture clusters, guided by learned spatial priors and semantic relationships extracted from populated scene datasets. The second stage utilizes an autoregressive transformer to complete scene synthesis by populating each identified cluster with contextually appropriate furniture arrangements. Our agentic AI framework shows autonomous reasoning capabilities through interpretable decision-making processes, contrasting with traditional black-box approaches. The method incorporates semantic-spatial consistency scoring, object occurrence heatmaps, and functional relationship modeling to ensure realistic and aesthetically pleasing layouts. We performed comprehensive qualitative and quantitative evaluations against recent methods, and our method delivers competitive performance while providing enhanced spatial reasoning and layout utilization. Our method addresses the key challenge of generating functional and semantically meaningful room settings, key to many intelligent environment applications in metaverse, extended reality, and personalized interior design systems.
Overview of our proposed framework. Our method for space-aware furniture recommendation and placement consists of scene parsing, rule engine-based prediction, and finally, autoregressive transformer modeling steps.