COGNIA: A Cognitive Collaboration Framework for Adaptive Decision-Making in Autonomous Embodied Intelligence

Abstract

Autonomous embodied intelligence has emerged as a promising new paradigm to enable intelligent agents to perceive, reason, and interact with dynamic environments. Recent advances in large language models (LLMs), vision-language models, and multi-agent systems have greatly improved embodied decision-making. But many current approaches work by treating perception, reasoning, memory, and cooperation separately. As an effect, the model adaptation is relatively minimal and long term planning generally unwise. To meet these challenges, we propose COGNIA, a Cognitive Collaboration Framework for Adaptive Decision-Making in Autonomous Embodied Intelligence. Combining adaptive context representation, hierarchical cognitive memory, collaborative reasoning, decision optimization, explainable action planning, and self-reflection together, the proposed framework constitutes a powerful cognitive architecture. Additionally, a new Cognitive Collaboration Score (CCS) is proposed to assess candidate decisions jointly based on contextual relevance, memory consistency, reasoning confidence, and execution feasibility, allowing firm and contextually sensitive choice of actions. This framework was previously studied across several embodied AI benchmarks, as well as benchmarking against state-of-the-art solutions (ReAct, AutoGen, MetaGPT, RT-2, PaLM-E and Voyager). Experimental data show that COGNIA achieves a Task Success Rate of 97.48%, Planning Accuracy of 97.91%, Precision of 97.62%, Recall of 97.34%, F1-score of 97.48%, and Collaboration Efficiency of 96.84%, with an average planning time of 2.91 s. Ablation and scalability analysis additionally validate the use of the proposed cognitive modules and validate the robustness and scalability of the approach for solving difficult tasks in embodied intelligence.