arXiv subjects

Deng Liang

Publications and source records attributed to Deng Liang.

4 recordsLinked to original sources

arXiv2026

Online Evolution Strategy for Flow-Matching VLA Policies via Self-Supervised Trajectory Distribution Optimization

Vision-Language-Action (VLA) models based on generative frameworks, such as Flow Matching, have recently achieved impressive performance in robotic manipulation. Unlike deterministic policies, Flow Matching enables VLA models to learn conditional action trajectory distributions, where latent noise vectors induce different actions under the same task scenario. However, we observe that these distributions are often ill-formed, with successful and failed behaviors coexisting while considerable probability mass remains in unfavorable regions. To this end, we propose Online-ES, an online adaptation framework for Flow Matching VLAs based on Evolution Strategy (ES), which refines the learned action trajectory distribution through interaction feedback. Instead of pruning the latent noise space, our method performs evolutionary exploration directly in the action trajectory space, where diverse trajectories generated by Flow Matching provide candidate solutions for adaptation. By perturbing sampled trajectories and evaluating their execution outcomes, we derive a self-supervised MSE objective that transfers the evolution direction from trajectory space into model parameter space. Mathematically, we prove that the proposed objective provides an unbiased estimator of the optimal evolution direction. Moreover, we also incorporate failure experiences as negative feedback to regularize the evolution direction, steering the policy away from previously explored failure regions. Experiments in both simulation and real-world environments demonstrate that Online-ES achieves policy improvement comparable to reinforcement fine-tuning, without learning a value model or computing advantages.

cs.RO↗

arXiv2026

TAO-DA: Towards Autonomous Operation--A Dual-Arm Vision-Language-Action Model for Coordinated Manipulation

Vision-Language-Action (VLA) models provide a unified framework for grounding high-level semantic information into low-level robot actions, enabling scalable robotic manipulation across diverse tasks. However, existing VLA models lack explicit mechanisms to disentangle the states and intents of the two arms, leading to unintended cross-arm interference that degrades task execution success. To address this issue, we propose a symmetric Dual-Arm Expert (DAE) architecture built upon a shared Vision-Language Model (VLM) backbone with decoupled, arm-specific expert towers. Expert selection is carried out through a two-stage dual-arm intent routing scheme, in which experts are routed either by explicit language instructions in the first stage or by implicit visual semantics in the second stage. Moreover, we introduce a lightweight task progress prediction module that leverages cross-attention between the pre-chunk temporal features and semantic representations of proprioceptive and visual observations to accurately estimate frame-wise task completion progress. This module facilitates task progress synchronization to support coordinated scheduling for collaborative multi-robot tasks. Experimental results demonstrate the effectiveness of our model in dual-arm intent routing and the disentanglement of cross-arm interference, and further provide preliminary evidence of emergent skill generalization from single- to dual-arm tasks (as well as the reverse), together with cross-arm motion-domain skill transfer.

cs.RO↗

arXiv2025

Retrieval Backward Attention without Additional Training: Enhance Embeddings of Large Language Models via Repetition

Language models can be viewed as functions that embed text into Euclidean space, where the quality of the embedding vectors directly determines model performance, training such neural networks involves various uncertainties. This paper focuses on improving the performance of pre-trained language models in zero-shot settings through a simple and easily implementable method. We propose a novel backward attention mechanism to enhance contextual information encoding. Evaluated on the Chinese Massive Text Embedding Benchmark (C-MTEB), our approach achieves significant improvements across multiple tasks, providing valuable insights for advancing zero-shot learning capabilities.

cs.CL↗

arXiv2020

MLBF-Net: A Multi-Lead-Branch Fusion Network for Multi-Class Arrhythmia Classification Using 12-Lead ECG

Automatic arrhythmia detection using 12-lead electrocardiogram (ECG) signal plays a critical role in early prevention and diagnosis of cardiovascular diseases. In the previous studies on automatic arrhythmia detection, most methods concatenated 12 leads of ECG into a matrix, and then input the matrix to a variety of feature extractors or deep neural networks for extracting useful information. Under such frameworks, these methods had the ability to extract comprehensive features (known as integrity) of 12-lead ECG since the information of each lead interacts with each other during training. However, the diverse lead-specific features (known as diversity) among 12 leads were neglected, causing inadequate information learning for 12-lead ECG. To maximize the information learning of multi-lead ECG, the information fusion of comprehensive features with integrity and lead-specific features with diversity should be taken into account. In this paper, we propose a novel Multi-Lead-Branch Fusion Network (MLBF-Net) architecture for arrhythmia classification by integrating multi-loss optimization to jointly learning diversity and integrity of multi-lead ECG. MLBF-Net is composed of three components: 1) multiple lead-specific branches for learning the diversity of multi-lead ECG; 2) cross-lead features fusion by concatenating the output feature maps of all branches for learning the integrity of multi-lead ECG; 3) multi-loss co-optimization for all the individual branches and the concatenated network. We demonstrate our MLBF-Net on China Physiological Signal Challenge 2018 which is an open 12-lead ECG dataset. The experimental results show that MLBF-Net obtains an average $F_1$ score of 0.855, reaching the highest arrhythmia classification performance. The proposed method provides a promising solution for multi-lead ECG analysis from an information fusion perspective.

eess.SP↗