arXiv subjects

Zilong Xie

Publications and source records attributed to Zilong Xie.

6 recordsLinked to original sources

arXiv2026

Zero-Shot Robotic Manipulation via 3D Gaussian Splatting-Enhanced Multimodal Retrieval-Augmented Generation

Existing end-to-end approaches of robotic manipulation often lack generalization to unseen objects or tasks due to limited data and poor interpretability. While recent Multimodal Large Language Models (MLLMs) demonstrate strong commonsense reasoning, they struggle with geometric and spatial understanding required for pose prediction. In this paper, we propose RobMRAG, a 3D Gaussian Splatting-Enhanced Multimodal Retrieval-Augmented Generation (MRAG) framework for zero-shot robotic manipulation. Specifically, we construct a multi-source manipulation knowledge base containing object contact frames, task completion frames, and pose parameters. During inference, a Hierarchical Multimodal Retrieval module first employs a three-priority hybrid retrieval strategy to find task-relevant object prototypes, then selects the geometrically closest reference example based on pixel-level similarity and Instance Matching Distance (IMD). We further introduce a 3D-Aware Pose Refinement module based on 3D Gaussian Splatting into the MRAG framework, which aligns the pose of the reference object to the target object in 3D space. The aligned results are reprojected onto the image plane and used as input to the MLLM to enhance the generation of the final pose parameters. Extensive experiments show that on a test set containing 30 categories of household objects, our method improves the success rate by 7.76% compared to the best-performing zero-shot baseline under the same setting, and by 6.54% compared to the state-of-the-art supervised baseline. Our results validate that RobMRAG effectively bridges the gap between high-level semantic reasoning and low-level geometric execution, enabling robotic systems that generalize to unseen objects while remaining inherently interpretable.

cs.RO↗

arXiv2026

Learning to Rank Critical Road Segments via Heterogeneous Graphs with Origin-Destination Flow Integration

Existing learning-to-rank methods for road networks often fail to incorporate origin-destination (OD) flows and route information, limiting their ability to model long-range spatial dependencies. To address this gap, we propose HetGL2R, a heterogeneous graph learning framework for ranking road-segment importance. HetGL2R builds a tripartite graph that unifies OD flows, routes, and network topology, and further introduces attribute-guided graphs that elevate node attributes into explicit nodes to model functional similarity. A heterogeneous joint random walk algorithm (HetGWalk) jointly samples both graph types to generate context-rich node sequences. These sequences are encoded using a Transformer to learn embeddings that capture long-range structural dependencies induced by OD flows and route configurations, as well as functional associations derived from attribute similarity. Finally, a listwise ranking strategy with a KL-divergence loss evaluates and ranks segment importance. Experiments on three SUMO-generated simulated networks of different scales show that, against state-of-the-art methods, HetGL2R achieves average improvements of approximately 7.52%, 4.40% and 3.57% in ranking performance.

cs.LG↗

arXiv2025

E2E-AFG: An End-to-End Model with Adaptive Filtering for Retrieval-Augmented Generation

Retrieval-augmented generation methods often neglect the quality of content retrieved from external knowledge bases, resulting in irrelevant information or potential misinformation that negatively affects the generation results of large language models. In this paper, we propose an end-to-end model with adaptive filtering for retrieval-augmented generation (E2E-AFG), which integrates answer existence judgment and text generation into a single end-to-end framework. This enables the model to focus more effectively on relevant content while reducing the influence of irrelevant information and generating accurate answers. We evaluate E2E-AFG on six representative knowledge-intensive language datasets, and the results show that it consistently outperforms baseline models across all tasks, demonstrating the effectiveness and robustness of the proposed approach.

cs.CL↗

arXiv2025

From Pixels to Tokens: Byte-Pair Encoding on Quantized Visual Modalities

Multimodal Large Language Models have made significant strides in integrating visual and textual information, yet they often struggle with effectively aligning these modalities. We introduce a novel image tokenizer that bridges this gap by applying the principle of Byte-Pair Encoding (BPE) to visual data. Unlike conventional approaches that rely on separate visual encoders, our method directly incorporates structural prior information into image tokens, mirroring the successful tokenization strategies used in text-only Large Language Models. This innovative approach enables Transformer models to more effectively learn and reason across modalities. Through theoretical analysis and extensive experiments, we demonstrate that our BPE Image Tokenizer significantly enhances MLLMs' multimodal understanding capabilities, even with limited training data. Leveraging this method, we develop Being-VL-0, a model that demonstrates superior performance across various benchmarks and shows promising scalability, potentially paving the way for more efficient and capable multimodal foundation models.

cs.AI↗

arXiv2024

Hierarchical Spatial Proximity Reasoning for Vision-and-Language Navigation

Most Vision-and-Language Navigation (VLN) algorithms are prone to making inaccurate decisions due to their lack of visual common sense and limited reasoning capabilities. To address this issue, we propose a Hierarchical Spatial Proximity Reasoning (HSPR) method. First, we introduce a scene understanding auxiliary task to help the agent build a knowledge base of hierarchical spatial proximity. This task utilizes panoramic views and object features to identify types of nodes and uncover the adjacency relationships between nodes, objects, and between nodes and objects. Second, we propose a multi-step reasoning navigation algorithm based on the hierarchical spatial proximity knowledge base, which continuously plans feasible paths to enhance exploration efficiency. Third, we introduce a residual fusion method to improve navigation decision accuracy. Finally, we validate our approach with experiments on publicly available datasets including REVERIE, SOON, R2R, and R4R. Our code is available at https://github.com/iCityLab/HSPR

cs.CV↗

arXiv2023

Variable Selection in Latent Regression IRT Models via Knockoffs: An Application to International Large-scale Assessment in Education

International large-scale assessments (ILSAs) play an important role in educational research and policy making. They collect valuable data on education quality and performance development across many education systems, giving countries the opportunity to share techniques, organizational structures, and policies that have proven efficient and successful. To gain insights from ILSA data, we identify non-cognitive variables associated with students' academic performance. This problem has three analytical challenges: 1) academic performance is measured by cognitive items under a matrix sampling design; 2) there are many missing values in the non-cognitive variables; and 3) multiple comparisons due to a large number of non-cognitive variables. We consider an application to the Programme for International Student Assessment (PISA), aiming to identify non-cognitive variables associated with students' performance in science. We formulate it as a variable selection problem under a general latent variable model framework and further propose a knockoff method that conducts variable selection with a controlled error rate for false selections.

stat.ME↗