Mimir: A Neuro-Symbolic Memory System with Dynamic Grounding for Embodied Agents in Interactive Environments

2026-08-05

AI Summary

This work addresses the challenge faced by embodied agents in partially observable environments, where explicitly linking world states to long-term goals remains difficult. To this end, the authors propose Mimir, a neuro-symbolic memory system that decouples world memory from task memory and introduces a dynamic grounding mechanism. This mechanism explicitly binds the current goal to environmental facts before each action, thereby integrating structured knowledge with task progress to enable goal-directed planning and decision-making. Mimir employs a modular architecture that combines neural networks with symbolic representations. Evaluated on the EB-ALFRED and EB-Habitat benchmarks, Mimir achieves an average performance improvement of 23.0% (up to 42.5%), surpasses the state-of-the-art by 8.5% in overall success rate, and attains 86.0% success on the long-horizon subset of EB-Habitat.

Abstract

Long-horizon embodied task requires agents to act under partial observability while preserving both scene belief and execution progress. Flat histories or implicit policy states may contain past observations, but they do not provide an explicit interface for deciding which world facts support the currently active goal. We introduce Mimir, a neuro-symbolic memory that separates world memory from task memory and dynamically grounds them before each action. World memory maintains object locations, object states, and perceptual evidence, while task memory maintains an ordered goal agenda, progress state, hand state, failures, and execution constraints. A grounding module binds the active goal to recalled world candidates, fills missing source locations, and attaches evidence before planning and embodiment-specific execution. Across tested backbones, Mimir consistently improves on different EB-ALFRED and EB-Habitat tasks, with maximum gains of 42.5% and average gains of 23.0%, respectively. Compared with the best results among prior agent and memory systems evaluated under the same backbone, Mimir improves the overall average success rate by 8.5%. Finally, on the EB-Habitat Long-horizon subset, Mimir achieves 86.0% success rate, substantially outperforming current closed-source models. Our code will be released soon.

Problem

• embodied agents
• long-horizon tasks
• partial observability
• scene belief
• goal grounding

Innovation

• neuro-symbolic memory
• dynamic grounding
• embodied agents
• task memory
• world memory