Search:
Match:
3 results

Unified Embodied VLM Reasoning for Robotic Action

Published:Dec 30, 2025 10:18
1 min read
ArXiv

Analysis

This paper addresses the challenge of creating general-purpose robotic systems by focusing on the interplay between reasoning and precise action execution. It introduces a new benchmark (ERIQ) to evaluate embodied reasoning and proposes a novel action tokenizer (FACT) to bridge the gap between reasoning and execution. The work's significance lies in its attempt to decouple and quantitatively assess the bottlenecks in Vision-Language-Action (VLA) models, offering a principled framework for improving robotic manipulation.
Reference

The paper introduces Embodied Reasoning Intelligence Quotient (ERIQ), a large-scale embodied reasoning benchmark in robotic manipulation, and FACT, a flow-matching-based action tokenizer.

Analysis

This paper addresses the challenge of speech synthesis for the endangered Manchu language, which faces data scarcity and complex agglutination. The proposed ManchuTTS model introduces innovative techniques like a hierarchical text representation, cross-modal attention, flow-matching Transformer, and hierarchical contrastive loss to overcome these challenges. The creation of a dedicated dataset and data augmentation further contribute to the model's effectiveness. The results, including a high MOS score and significant improvements in agglutinative word pronunciation and prosodic naturalness, demonstrate the paper's significant contribution to the field of low-resource speech synthesis and language preservation.
Reference

ManchuTTS attains a MOS of 4.52 using a 5.2-hour training subset...outperforming all baseline models by a notable margin.

Research#Image Generation📝 BlogAnalyzed: Dec 29, 2025 01:43

Just Image Transformer: Flow Matching Model Predicting Real Images in Pixel Space

Published:Dec 14, 2025 07:17
1 min read
Zenn DL

Analysis

The article introduces the Just Image Transformer (JiT), a flow-matching model designed to predict real images directly within the pixel space, bypassing the use of Variational Autoencoders (VAEs). The core innovation lies in predicting the real image (x-pred) instead of the velocity (v), achieving superior performance. The loss function, however, is calculated using the velocity (v-loss) derived from the real image (x) and a noisy image (z). The article highlights the shift from U-Net-based models, prevalent in diffusion-based image generation like Stable Diffusion, and hints at further developments.
Reference

JiT (Just image Transformer) does not use VAE and performs flow-matching in pixel space. The model performs better by predicting the real image x (x-pred) rather than the velocity v.