DeepHA: Scaling Action Chains Elicits Deep Hierarchical Agents

Prevailing autonomous agents are often constrained by a single, predefined action space, which limits their generalization capabilities across diverse tasks and can introduce compounding errors through decoupled policy execution. To address these limitations, we introduce the Deep Hierarchical Agent (DeepHA), a unified architecture that operates across a mixture of heterogeneous action spaces, flexibly generating actions ranging from high-level semantic skills to low-level motor controls. We further propose a Chain-of-Action (CoA) reasoning framework, which enables the agent to use higher-level abstract actions as structured `thoughts' to guide the generation of more granular, subsequent actions. To manage the computational demands of this deep reasoning in long-horizon tasks, we develop a memory-efficient mechanism that dynamically compresses historical context and leverages Key-Value (KV) caching, reducing context length by approximately 75% without sacrificing performance. We conduct extensive evaluations on a new, large-scale benchmark of over 800 diverse Minecraft tasks. Results show that DHA significantly outperforms prior methods, establishing a new state-of-the-art and demonstrating superior generalization, particularly in complex, multi-step planning tasks. Our work presents a novel, unified framework for building more capable and efficient autonomous agents.

Currently, most intelligent AI can only act in predefined ways, finding it difficult to handle diverse tasks and prone to consecutive errors. We have designed a new intelligent agent architecture that allows AI to flexibly switch between different action modes, both understanding macro-level thinking and completing detailed operations. At the same time, we have paired it with a dedicated reasoning method, making AI think like humans, first sorting out the思路 before taking step-by-step actions, and have optimized the operation mechanism, significantly reducing computational costs without affecting the results. In tests of over eight hundred game tasks, this solution outperforms existing technology, especially excelling in complex multi-step tasks, providing a new direction for creating stronger and more efficient intelligent AI.