OpenHA: A Series of Open-Source Hierarchical Agentic Models in Minecraft

2025-09-12

AI Summary

In open-world environments, systematic approaches to action-space abstraction and tokenization for Vision-Language-Action (VLA) agents remain lacking. This paper addresses this gap for Minecraft-based VLA agents by proposing a unified hierarchical agent framework centered on the “Chain of Action” paradigm: abstract actions are modeled as learnable reasoning steps—not predefined sub-policy instructions—enabling end-to-end integration of high-level planning and low-level execution. Methodologically, we integrate multi-granularity action-space modeling, task-aware tokenization, and joint large-scale behavioral cloning. Evaluated on a benchmark of 800+ diverse tasks, our All-in-One agent significantly outperforms specialized baselines, establishing new state-of-the-art performance. We publicly release the full model, toolchain, and dataset to foster reproducible research and community advancement.

Technology Category

• Multiagent Systems: Adversarial Agents
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Cognitive Modeling & Cognitive Systems: Agent Architectures

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Agentic search

Abstract

The choice of action spaces is a critical yet unresolved challenge in developing capable, end-to-end trainable agents. This paper first presents a large-scale, systematic comparison of prominent abstracted action spaces and tokenizers for Vision-Language-Action (VLA) or hierarchical agent models in the open-ended Minecraft. Our analysis reveals that no single action space is universally optimal; instead, the most effective abstraction is highly task-dependent, creating a dilemma for building generalist agents. To resolve this, we introduce Chain of Action (CoA), a novel framework that unifies high-level planning and low-level control within a single, monolithic VLA model. CoA treats an abstracted action not as a command for a separate policy, but as an intermediate reasoning step--akin to a chain of thought--that guides the generation of the final, executable action. Furthermore, we demonstrate that an All-in-One agent trained on a diverse mixture of action spaces using the CoA paradigm learns a more robust and generalizable policy. This unified agent achieves a new state-of-the-art, improving the overall task success rate over strong, specialized baselines. To foster reproducible research, we release the OpenHA (Open Hierarchical Agents) suite, which includes our comprehensive benchmark of over 800 distinct tasks, curated datasets, source code, and all pretrained model checkpoints at https://github.com/CraftJarvis/OpenHA

Problem

• Comparing action spaces and tokenizers for Minecraft agents
• Resolving task-dependent optimality in generalist agent design
• Unifying high-level planning and low-level control in VLA models

Innovation

• Unified CoA framework for hierarchical planning
• Treats actions as intermediate reasoning steps
• Trains on diverse action spaces mixture