Cognitive Architectures for Adaptive Mission Planning in Heterogeneous Unmanned Systems: A Comparative Study

Abstract

Heterogeneous unmanned systems comprising aerial, ground, surface, and underwater vehicles require sophisticated cognitive architectures to enable adaptive mission planning in dynamic, uncertain operational environments. This comparative study systematically analyzes prominent cognitive architectures—SOAR, ACT-R, CLARION, LIDA, and Sigma—evaluating their suitability for autonomous mission management in multi-domain unmanned operations. We examine the fundamental cognitive capabilities required for adaptive planning including perception-action cycles, working and long-term memory structures, learning mechanisms, goal management hierarchies, and metacognitive reasoning. The study establishes a comprehensive evaluation framework encompassing architectural modularity, computational scalability, real-time performance, learning flexibility, and integration compatibility with existing unmanned system control stacks. Through detailed analysis, we assess how each architecture handles critical mission planning challenges: dynamic task allocation among heterogeneous agents, replanning under resource constraints, handling of incomplete information, and adaptation to adversarial actions. Particular attention is devoted to hybrid architectures that combine symbolic reasoning with subsymbolic learning, examining their potential to bridge the gap between high-level mission objectives and low-level vehicle control. We investigate the incorporation of modern AI techniques including deep learning for perception, reinforcement learning for decision optimization, and large language models for natural language mission specification within traditional cognitive frameworks. The comparative analysis reveals that no single architecture dominates across all evaluation dimensions, with SOAR demonstrating superior symbolic reasoning capabilities while hybrid approaches show promise for complex, data-rich environments. The study concludes by proposing design principles for next-generation cognitive architectures specifically tailored to unmanned systems, emphasizing modularity, explainability, and graceful degradation under uncertainty.