Author name cluster

Guanghui Liu

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

6 papers

2 author rows

EAAI Journal 2026 Journal Article

Seeing the unseen: Semantic segmentation and uncertainty quantification for delamination detection in building facades

Yuebo Meng
Guotong Yin
Songtao Ye
Qiaoqiao Wang
Guanghui Liu
Xiaohan Li
Xiaojiao Geng

Accurate detection of delamination in building facades is critical for prolonging service life and ensuring structural safety. Current inspection methodologies heavily rely on manual interpretation, lacking efficiency and intelligent robustness. While infrared thermography provides a non-destructive means for detecting subsurface delamination, its accuracy is often compromised by low thermal contrast under uncontrolled conditions and the absence of uncertainty quantification in deep learning models. To address these limitations, this paper proposes TIHSNet, a novel delamination detection framework based on semantic segmentation and uncertainty quantification. Specifically, a physics-informed thermal gradient attention module is introduced to emphasize thermodynamically meaningful gradients and enable accurate delamination boundary delineation. Subsequently, a dual output mechanism is proposed to simultaneously generate prediction and uncertainty maps, enabling quantitative assessment of predictive reliability and identification of regions requiring expert review. To further enhance spatial localization, visible light images are integrated to capture tile boundary information and support spatial classification of delamination. Experiments were conducted on a self constructed dataset comprising 2102 infrared thermography and visible light images collected from reinforced concrete and brick masonry walls. The results demonstrate that TIHSNet achieves a precision of 96. 1%, surpassing traditional thresholding methods with a 27. 9% gain, and further outperforming existing deep learning approaches by 10. 5%. The uncertainty quantification results further validate the model’s robustness and its ability to support reliable decision making in real world inspection scenarios.

Details DOI

IROS Conference 2025 Conference Paper

A stepwise identification framework for determining the physical feasibility parameters of robot dynamics

Guanghui Liu
Fuyuan Tan
Lijin Fang
Hualiang Zhang
Qiang Li

This paper introduces a systematic approach to identifying a physically feasible set of robot dynamics parameters. The framework consists of four steps: 1) Identification of robot dynamics parameters using least squares combined with a linear friction model. 2) Construction of a weighting matrix based on the least squares identification error, and performing weighted least squares identification combined with the linear friction model. 3) Introduction of a nonlinear friction model to fit joint friction. 4) Optimization of the remaining robot dynamics parameters to adhere to physical feasibility constraints. Various combinations of identification methods with linear or nonlinear friction models are analyzed experimentally, using a 6-DoF industrial robot and a 7-DoF collaborative robot, respectively, to demonstrate the effectiveness of the proposed recognition framework. Experimental results affirm that the proposed method provides accurate estimates of the robot joint torques while maintaining the physical feasibility of the dynamics.

Details

AAAI Conference 2025 Conference Paper

FlowPolicy: Enabling Fast and Robust 3D Flow-Based Policy via Consistency Flow Matching for Robot Manipulation

Qinglun Zhang
Zhen Liu
Haoqiang Fan
Guanghui Liu
Bing Zeng
Shuaicheng Liu

Robots can acquire complex manipulation skills by learning policies from expert demonstrations, which is often known as vision-based imitation learning. Generating policies based on diffusion and flow matching models has been shown to be effective, particularly in robotic manipulation tasks. However, recursion-based approaches are inference inefficient in working from noise distributions to policy distributions, posing a challenging trade-off between efficiency and quality. This motivates us to propose FlowPolicy, a novel framework for fast policy generation based on consistency flow matching and 3D vision. Our approach refines the flow dynamics by normalizing the self-consistency of the velocity field, enabling the model to derive task execution policies in a single inference step. Specifically, FlowPolicy conditions on the observed 3D point cloud, where consistency flow matching directly defines straight-line flows from different time states to the same action space, while simultaneously constraining their velocity values, that is, we approximate the trajectories from noise to robot actions by normalizing the self-consistency of the velocity field within the action space, thus improving the inference efficiency. We validate the effectiveness of FlowPolicy in Adroit and Metaworld, demonstrating a 7× increase in inference speed while maintaining competitive average success rates compared to state-of-the-art methods.

PDF Details DOI

EAAI Journal 2024 Journal Article

Geometry perception and motion planning in robotic assembly based on semantic segmentation and point clouds reconstruction

Yuze Jiang
Guanghui Liu
Zhouzhou Huang
Bin Yang
Wenyu Yang

The assembly operations for robots in unstructured environments face tasks with generalized geometric information. The most essential issue for geometry perception is to estimate the relative pose between mating parts and then plan the corresponding actions. Current perception methods mainly focus on template matching of specific parts, which is still lacking in geometric generalization. To enhance the perception generalization ability, we propose an operation framework combining geometry perception and motion planning for robotic assembly manipulation. Semantic segmentation embodies a strong generalization capacity in image processing, which we utilized to extract the geometric features of assembly parts. The networks are trained via a self-made peg-hole image dataset. To reduce the measurement noise, the virtual point clouds for the peg and hole section are reconstructed with semantic mask and camera image principle, and a uniformization algorithm is adopted to raise the quality of the point clouds. The registration is utilized with the noise-free, uniform virtual point clouds, and the relative pose between two mating parts can be estimated more precisely. Also, the sequence of interactive geometry perception actions is defined. The performance of our framework is validated on assembly experiments, including different geometries in and out of our dataset. The results show that the framework proposed in this paper can perceive the generalization of geometry features for assembly parts, and leads to intelligent robotic assembly.

Details DOI

AAAI Conference 2024 Conference Paper

SpectralNeRF: Physically Based Spectral Rendering with Neural Radiance Field

Ru Li
Jia Liu
Guanghui Liu
Shengping Zhang
Bing Zeng
Shuaicheng Liu

In this paper, we propose SpectralNeRF, an end-to-end Neural Radiance Field (NeRF)-based architecture for high-quality physically based rendering from a novel spectral perspective. We modify the classical spectral rendering into two main steps, 1) the generation of a series of spectrum maps spanning different wavelengths, 2) the combination of these spectrum maps for the RGB output. Our SpectralNeRF follows these two steps through the proposed multi-layer perceptron (MLP)-based architecture (SpectralMLP) and Spectrum Attention UNet (SAUNet). Given the ray origin and the ray direction, the SpectralMLP constructs the spectral radiance field to obtain spectrum maps of novel views, which are then sent to the SAUNet to produce RGB images of white-light illumination. Applying NeRF to build up the spectral rendering is a more physically-based way from the perspective of ray-tracing. Further, the spectral radiance fields decompose difficult scenes and improve the performance of NeRF-based methods. Comprehensive experimental results demonstrate the proposed SpectralNeRF is superior to recent NeRF-based methods when synthesizing new views on synthetic and real datasets. The codes and datasets are available at https://github.com/liru0126/SpectralNeRF.

PDF Details DOI

AAAI Conference 2022 Conference Paper

FINet: Dual Branches Feature Interaction for Partial-to-Partial Point Cloud Registration

Hao Xu
Nianjin Ye
Guanghui Liu
Bing Zeng
Shuaicheng Liu

Data association is important in the point cloud registration. In this work, we propose to solve the partial-to-partial registration from a new perspective, by introducing multi-level feature interactions between the source and the reference clouds at the feature extraction stage, such that the registration can be realized without the attentions or explicit mask estimation for the overlapping detection as adopted previously. Specifically, we present FINet, a feature interactionbased structure with the capability to enable and strengthen the information associating between the inputs at multiple stages. To achieve this, we first split the features into two components, one for rotation and one for translation, based on the fact that they belong to different solution spaces, yielding a dual branches structure. Second, we insert several interaction modules at the feature extractor for the data association. Third, we propose a transformation sensitivity loss to obtain rotation-attentive and translation-attentive features. Experiments demonstrate that our method performs higher precision and robustness compared to the state-of-the-art traditional and learning-based methods. Code is available at https: //github. com/megvii-research/FINet.

PDF Details