Hierarchical Neuro-Symbolic Decision Transformer

2025-03-10

AI Summary

This work addresses the interpretability and reliability challenges in long-horizon decision-making under high-dimensional uncertainty. We propose a hierarchical neuro-symbolic decision framework: a high-level logical symbolic planner generates verifiable action sequences satisfying global constraints, while a low-level decision Transformer, conditioned on subgoal tokens derived from those actions, produces fine-grained motor commands. To our knowledge, this is the first end-to-end integration of symbolic planning with decision Transformers. We introduce error propagation analysis to formally guarantee inter-layer reliability and design a joint neuro-symbolic training scheme with subgoal-conditional modeling. Evaluated on a multi-key–lock–door–object-collection grid world, our approach significantly outperforms end-to-end neural baselines in task success rate and policy efficiency, while providing strong interpretability, formal verifiability, and zero-shot generalization capability.

Technology Category

• Machine Learning: Neuro-Symbolic Learning
• Reasoning under Uncertainty: Sequential Decision Making
• Humans and AI: Planning and Decision Support for Human-Machine Teams

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

We present a hierarchical neuro-symbolic control framework that couples classical symbolic planning with transformer-based policies to address complex, long-horizon decision-making tasks. At the high level, a symbolic planner constructs an interpretable sequence of operators based on logical propositions, ensuring systematic adherence to global constraints and goals. At the low level, each symbolic operator is translated into a sub-goal token that conditions a decision transformer to generate a fine-grained sequence of actions in uncertain, high-dimensional environments. We provide theoretical analysis showing how approximation errors from both the symbolic planner and the neural execution layer accumulate. Empirical evaluations in grid-worlds with multiple keys, locked doors, and item-collection tasks show that our hierarchical approach outperforms purely end-to-end neural approach in success rates and policy efficiency.

Problem

• Addresses complex long-horizon decision-making tasks
• Combines symbolic planning with transformer-based policies
• Improves success rates and policy efficiency in grid-worlds

Innovation

• Hierarchical neuro-symbolic control framework
• Symbolic planner with transformer-based policies
• Sub-goal tokens for fine-grained action sequences