Skip to content
Open access

CARTA: Context-Aware Dual Retrieval and Chain-of-Thought Task Allocation for Environment-Grounded Elderly Care

Aug 2026 · Vietnam Journal of Computer Science · pp. 1-40 · 0 citations

TL;DR

The results establish that explicit environmental grounding and deliberative multi-agent reasoning are essential for safe, contextually appropriate LLM-based elderly care.

Abstract

Enabling elderly individuals to age independently at home requires intelligent assistive systems that can understand complex care needs and coordinate appropriate responses. While large language models (LLMs) show promise for adaptive assistance, current eldercare systems suffer from critical limitations: they generate physically infeasible actions due to weak environmental grounding, cannot coordinate multi-faceted care needs through single-agent processing, and lack safety mechanisms to prevent hallucination of non-existent resources. We present CARTA (Context-Aware Dual Retrieval and Chain-of-Thought Task Allocation), a framework that addresses these gaps through two innovations. First, our dual-path retrieval architecture combines semantic knowledge retrieval with graph-constrained retrieval-augmented reasoning, constructing an explicit environment graph encoding object availability, functional affordances, and spatial relationships. This grounds LLM reasoning in physical reality, improving the environmental executability of generated plans by constraining them to objects and affordances actually present in the scene. Second, our Chain-of-Thought multi-agent coordination enables specialized agents to deliberatively decompose ambiguous requests, negotiate responsibilities based on capabilities and constraints, and execute interventions concurrently. Experimental evaluation demonstrates substantial improvements: CARTA achieves 81.8% environment constraint satisfaction versus 65.5% for BM25 and 76.4% for dense retrieval, preventing resource hallucination. On multi-agent task execution, CARTA attains 78.1% success rate on complex tasks and 75.0% on vague commands-a 7.9% improvement over state-of-the-art planners on ambiguous scenarios requiring adaptive coordination. These results establish that explicit environmental grounding and deliberative multi-agent reasoning are essential for safe, contextually appropriate LLM-based elderly care.

Read PDF

Similar papers

Preprint Nov 2026

Embodied RAG++: Retrieval-Augmented Generation From What Exists to What Works in Embodied Scenes

To serve a request in a human environment, a robot system must find an object that satisfies a functional requirement and a robot that can feasibly reach it. Scene graphs record which objects exist and where, but they often omit capabilities documented only in product manuals. Retrieval-augmented generation methods can...

Fan Yang, Tatsuya Kikuzuki, Jun-Jie Liang et al. · 0 citations
Open access Oct 2026

DyGRAV: Dynamic Grounding and Relational Reasoning for Vision–Language Navigation

Vision–Language Navigation (VLN) requires embodied agents to interpret natural-language instructions in 3D environments. While significant progress has been made, state-of-the-art agents often struggle with instructions that demand fine-grained visual understanding, correctly grounding specific object attributes, disti...

Zain Chaudhary, Zohaib Chaudhary, Hammad Ahmed et al. · 0 citations
Sep 2026

Beyond Retrieval Relevance: Scene-Grounded Risk Entailment for Vision-Language Driving

Retrieval-augmented generation (RAG) gives vision--language driving systems access to external safety knowledge, yet a retrieved risk rule may be relevant without applying to the current scene. A vision--language model (VLM) receiving such knowledge must ground objects, bind entities across time, and verify relations b...

Jia-Xin Liu, Rui-Lin Yu, Liang Peng et al. · 0 citations
#artificial intelligence Preprint Oct 2026

OntoAR-HAR: Ontology-Guided Agentic RAG with Personalized Memory for Human Activity Recognition

LLMs increasingly support sensor-based human activity recognition (HAR) by linking motion signals with activity semantics and physical knowledge. Retrieval-augmented generation (RAG) grounds decisions in sensor examples and domain features, enabling training-free target-user inference with less dependence on raw numeri...

Xiang-Qing Lu, Tao Zhu, Liang Yu et al. · 0 citations
Preprint Oct 2026

OntoPlan: An Ontology-Grounded Scene Representation and Agentic Framework for Scalable Robot Task Planning

Large language model (LLM)-based robot task planning is promising for open-ended instruction following, but degrades on long-horizon tasks in large environments. When spatial information is conveyed to the LLM through text, the model can fail to capture spatial context, and token cost grows with environment size. Gener...

Hyeongwoo Nam, Woongje Cho, Juwon Kim et al. · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.