The rapid evolution of Embodied AI and Large Language Models presents significant opportunities for home robotics, yet challenges persist in enabling robots to execute long-term, high-level natural language instructions. Current LLM-driven embodied agents often suffer from sub-optimal task planning, limited memory systems struggling with multi-hop queries, and inflexible agent routing mechanisms. To address these limitations, we propose the Context-Rich Adaptive Embodied Agent (CRAEA) framework, designed to significantly enhance task planning and memory-augmented question answering in household robots. CRAEA integrates core components: Semantic-Enhanced Task Planning (SETP), which enriches LLM-driven planning with object relationship graphs, hierarchical strategies, and implicit physical constraints; Multi-Modal Contextual Memory (MMCM), which stores comprehensive contextual memory units in a relational graph for sophisticated multi-hop reasoning and employs an advanced retrieval mechanism with temporal decay; and Adaptive Agent Routing and Coordination (AARC), featuring intent recognition with confidence evaluation, proactive clarification, and a planning feedback loop. Evaluated in an artificial home environment across complex tidying scenarios, CRAEA consistently demonstrates superior performance. Empirical results show that CRAEA achieves notable improvements in Task Planning Accuracy, Knowledge Base Response Total Validity, and Agent Routing Success Rate compared to baseline methods. A human evaluation further confirms enhanced coherence, naturalness, and user satisfaction, while an ablation study validates the critical contribution of each proposed module. CRAEA represents a significant step towards more intelligent, robust, and user-adaptive home robots.
Speech Emotion Recognition (SER) faces significant generalization challenges, particularly in Cross-Linguistic SER (CLSER), due to linguistic and cultural variabilities. Existing approaches struggle with robustly fusing diverse features and adapting to cross-linguistic discrepancies. To address this, we propose the Adaptive Contextualized Multi-feature Fusion Network (ACMF-Net), a novel architecture built on a ``contextualize first, then adaptively fuse'' paradigm. ACMF-Net leverages HuBERT embeddings alongside contextualized Mel-frequency Cepstral Coefficients (MFCCs) and prosodic features, each processed by dedicated Transformer encoders to capture rich temporal dependencies. A core innovation is the Dynamic Gating mechanism, which intelligently learns to dynamically weight the contributions of these heterogeneous feature modalities based on the input speech characteristics, thereby enhancing robustness against cross-linguistic variations. Evaluated on the IEMOCAP dataset for source language performance, ACMF-Net achieved superior Unweighted Accuracy (UAR), outperforming strong baselines and existing multi-feature fusion models. Furthermore, through few-shot fine-tuning on diverse target languages, ACMF-Net consistently demonstrated superior cross-linguistic generalization. An ablation study confirmed the critical contribution of each proposed component, especially the Dynamic Gating mechanism. These results underscore ACMF-Net's potential to significantly advance robust and generalized emotion recognition across linguistic boundaries.
Recent advances in language models have greatly improved their ability to understand and generate natural language. Yet, when applied to specialized fields such as financial decision support or complex system diagnosis, they often struggle with limited domain expertise, weak logical reasoning, and unreliable performance under uncertainty. Fine-tuning these large models is typically constrained by cost, privacy, and proprietary limitations. To overcome these issues, this study introduces CRONUS: Contextual Reasoning Orchestration for Navigating Uncertain Scenarios, a framework designed to enhance general-purpose models in domain-specific and decision-intensive tasks. CRONUS employs a lightweight, trainable agent named CARA (Context-Aware Reasoning Agent) to guide the reasoning process of black-box models through structured contextual instructions. CARA is developed via a three-stage training strategy that builds domain understanding, refines reasoning path generation, and optimizes dynamic decision prompts. Experiments in financial analysis tasks show that CRONUS markedly improves reasoning depth, consistency, and robustness compared with direct model use, retrieval-augmented methods, and specialized domain models, demonstrating its effectiveness for high-stakes decision-making in complex environments.