Universal Skeleton Understanding via Differentiable Rendering and MLLMs

2026-03-18

AI Summary

This work addresses the challenge that multimodal large language models (MLLMs) struggle to directly process non-visual structured data, such as human skeletal sequences, with existing approaches often suffering from information loss or limited generalization. To overcome this, the authors propose SkeletonLLM—a general-purpose framework for skeleton understanding—that leverages DrAction, a differentiable and format-agnostic renderer, to map arbitrary skeleton sequences end-to-end into compact image sequences. This enables MLLMs to interpret skeletal data natively through visual perception. By integrating a synergistic training strategy combining causal reasoning distillation and discriminative fine-tuning, the model achieves significant improvements in structured reasoning and generalization across diverse tasks, including action recognition, caption generation, logical reasoning, and cross-format transfer, thereby offering an effective pathway to extend MLLMs to non-native modalities.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Computer Vision: Multi-modal Vision

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Multimodal large language models (MLLMs) exhibit strong visual-language reasoning, yet remain confined to their native modalities and cannot directly process structured, non-visual data such as human skeletons. Existing methods either compress skeleton dynamics into lossy feature vectors for text alignment, or quantize motion into discrete tokens that generalize poorly across heterogeneous skeleton formats. We present SkeletonLLM, which achieves universal skeleton understanding by translating arbitrary skeleton sequences into the MLLM's native visual modality. At its core is DrAction, a differentiable, format-agnostic renderer that converts skeletal kinematics into compact image sequences. Because the pipeline is end-to-end differentiable, MLLM gradients can directly guide the rendering to produce task-informative visual tokens. To further enhance reasoning capabilities, we introduce a cooperative training strategy: Causal Reasoning Distillation transfers structured, step-by-step reasoning from a teacher model, while Discriminative Finetuning sharpens decision boundaries between confusable actions. SkeletonLLM demonstrates strong generalization on diverse tasks including recognition, captioning, reasoning, and cross-format transfer -- suggesting a viable path for applying MLLMs to non-native modalities. Code will be released upon acceptance.

Problem

• skeleton understanding
• multimodal large language models
• non-visual data
• format heterogeneity
• structured data

Innovation

• Differentiable Rendering
• Multimodal Large Language Models
• Skeleton Understanding
• Causal Reasoning Distillation
• Format-Agnostic Representation