Open Access

ARTICLE

Neuro-Symbolic Reasoning for 3D Human Pose Analysis

Yucheng Huang^1,*, Xingyu Gao², Jianze Wei², Hong Yan¹

1 Department of Electrical Engineering, City University of Hong Kong, Hong Kong SAR, China
2 Institute of Microelectronics, Chinese Academy of Sciences, Beijing, China

* Corresponding Author: Yucheng Huang. Email:

Computer Modeling in Engineering & Sciences 2026, 148(3), 34 https://doi.org/10.32604/cmes.2026.088326

Received 01 July 2026; Accepted 02 September 2026; Issue published 28 September 2026

Abstract

Current 3D Human Pose and Shape estimation methods predominantly focus on numerical coordinate regression, treating pose analysis as a geometric mapping task rather than a structured reasoning problem. This reliance on quantitative output creates an interpretability gap, as these models fail to provide high-level, qualitative justifications for their predictions, such as why a specific pose is physically unstable or biomechanically incorrect. To bridge this gap, we propose a novel framework that leverages neuro-symbolic reasoning for the analysis of human poses from multimodal inputs. Our approach integrates symbolic expressions and logical deduction rules into a large language model-based reasoning pipeline. Specifically, the system first translates natural language descriptions of human interactions into structured symbolic representations, then derives a step-by-step plan to reason about 3D poses and verify natural language statements using symbolic logical rules. A verification mechanism ensures the reliability of both the symbolic translation and the reasoning process. Despite its simplicity, our approach captures the physical characteristics of pose and understands the pose structure via neuro-symbolic reasoning. Evaluations demonstrate that the integration of Large Language Models (LLMs) with neuro-symbolic logic provides a reliable and interpretable approach for automated posture balance assessment and anomaly identification.

---

Keywords

---