Coordinate-Based Dual-Constrained Autoregressive Motion Generation

2026-04-09

AI Summary

This work addresses error amplification in diffusion-based text-to-motion generation and mode collapse in autoregressive models caused by motion discretization. To overcome these limitations, the authors propose the first dual-constrained autoregressive generation framework operating on continuous motion coordinates. The approach integrates a diffusion-inspired multilayer perceptron to enhance motion fidelity and introduces a dual-constrained causal masking mechanism to guide the generation process. Furthermore, it employs joint text-motion encoding to achieve precise semantic alignment. Evaluated on a newly established benchmark, the proposed framework achieves state-of-the-art performance in both motion fidelity and text-motion semantic consistency, setting a new standard for text-driven motion generation and editing.

Abstract

Text-to-motion generation has attracted increasing attention in the research community recently, with potential applications in animation, virtual reality, robotics, and human-computer interaction. Diffusion and autoregressive models are two popular and parallel research directions for text-to-motion generation. However, diffusion models often suffer from error amplification during noise prediction, while autoregressive models exhibit mode collapse due to motion discretization. To address these limitations, we propose a flexible, high-fidelity, and semantically faithful text-to-motion framework, named Coordinate-based Dual-constrained Autoregressive Motion Generation (CDAMD). With motion coordinates as input, CDAMD follows the autoregressive paradigm and leverages diffusion-inspired multi-layer perceptrons to enhance the fidelity of predicted motions. Furthermore, a Dual-Constrained Causal Mask is introduced to guide autoregressive generation, where motion tokens act as priors and are concatenated with textual encodings. Since there is limited work on coordinate-based motion synthesis, we establish new benchmarks for both text-to-motion generation and motion editing. Experimental results demonstrate that our approach achieves state-of-the-art performance in terms of both fidelity and semantic consistency on these benchmarks.

Problem

• text-to-motion generation
• error amplification
• mode collapse
• motion discretization
• semantic consistency

Innovation

• coordinate-based motion generation
• autoregressive modeling
• dual-constrained causal mask
• text-to-motion synthesis
• motion fidelity