arXiv subjects

He Liang

Publications and source records attributed to He Liang.

12 recordsLinked to original sources

arXiv2026

MemGuard: Persisting Verifier Signals for LLM-Agent Memory Governance

LLM agents are moving from single-prompt use to long task streams in which reusable memory becomes a core capability for terminal, software-engineering, and web tasks. Such memory is useful only when stored experience remains reliable across hundreds of interactions, but two failure modes break that assumption in practice. The first is unreliable admission: failed trajectories,accidental successes, and misleading observations enter memory because they appear relevant, then mislead later decisions. The second is memory drift: long-running banks accumulate duplicate, stale, and conflicting records that retrieval alone cannot repair. MemGuard's key distinction is to treat verifier output not as a one-shot filter, but as persistent lifecycle metadata. It converts multi-criteria score-token verification into reward, confidence, label, and uncertainty descriptors that are attached to every candidate before activation and reused during retrieval, conflict resolution, summarization, and archival. We evaluate MemGuard on Terminal-Bench 2.0, SWE-Bench Verified, WebArena, and Mind2Web across four backbones, comparing against four memory baselines plus a verifier-only control under matched runtime budgets. Averaged over five seeds, MemGuard achieves the best success metric and lowest average steps in all 16 backbone-benchmark settings, improving over ReasoningBank, the strongest prior baseline among the memory methods we evaluate, with a largest gain of 7.9 success-rate points on WebArena, 5.6 step-success-rate points on Mind2Web, and 2.4-3.5 points on terminal and software-engineering benchmarks. Code is available at https://github.com/whyyyyy123/MemGuard.

cs.AI↗

arXiv2026

CAIRN: Cross-Room 3D Scene Understanding with Topology-Aware Large Multimodal Models

Existing 3D scene-grounded Large Language Models (3D-LLMs) focus on answering questions grounded in simplified single-room 3D scenes, lacking the ability to reason over real-world household environments containing multiple interconnected rooms and diverse object categories. We introduce CAIRN, a topology-aware 3D-LLM for multi-room 3D scene understanding. CAIRN aligns transformer attention with scene hierarchy, giving the model explicit awareness of object-level relations and room-level connectivity. It enriches object tokens with room-local relational context via a graph neural network, introduces learned room tokens for room-level abstraction, and applies a hierarchical attention mask with geometric bias to route information according to scene topology. CAIRN is developed on CAIRN-MR, a benchmark we introduce on HM3D for multi-room 3D scene understanding, covering grounding, captioning, and four question-answering tasks that progressively evaluate from intra-room perception to cross-room reasoning. Experiments show that CAIRN outperforms prior 3D-LLMs by a large margin across all CAIRN-MR tasks while remaining competitive on five single-room benchmarks.

cs.CV↗

arXiv2025

A Framework for Deploying Learning-based Quadruped Loco-Manipulation

Quadruped mobile manipulators offer strong potential for agile loco-manipulation but remain difficult to control and transfer reliably from simulation to reality. Reinforcement learning (RL) shows promise for whole-body control, yet most frameworks are proprietary and hard to reproduce on real hardware. We present an open pipeline for training, benchmarking, and deploying RL-based controllers on the Unitree B1 quadruped with a Z1 arm. The framework unifies sim-to-sim and sim-to-real transfer through ROS, re-implementing a policy trained in Isaac Gym, extending it to MuJoCo via a hardware abstraction layer, and deploying the same controller on physical hardware. Sim-to-sim experiments expose discrepancies between Isaac Gym and MuJoCo contact models that influence policy behavior, while real-world teleoperated object-picking trials show that coordinated whole-body control extends reach and improves manipulation over floating-base baselines. The pipeline provides a transparent, reproducible foundation for developing and analyzing RL-based loco-manipulation controllers and will be released open source to support future research.

cs.RO↗

arXiv2025

MoReGen: Multi-Agent Motion-Reasoning Engine for Code-based Text-to-Video Synthesis

While text-to-video (T2V) generation has achieved remarkable progress in photorealism, generating intent-aligned videos that faithfully obey physics principles remains a core challenge. In this work, we systematically study Newtonian motion-controlled text-to-video generation and evaluation, emphasizing physical precision and motion coherence. We introduce MoReGen, a motion-aware, physics-grounded T2V framework that integrates multi-agent LLMs, physics simulators, and renderers to generate reproducible, physically accurate videos from text prompts in the code domain. To quantitatively assess physical validity, we propose object-trajectory correspondence as a direct evaluation metric and present MoReSet, a benchmark of 1,275 human-annotated videos spanning nine classes of Newtonian phenomena with scene descriptions, spatiotemporal relations, and ground-truth trajectories. Using MoReSet, we conduct experiments on existing T2V models, evaluating their physical validity through both our MoRe metrics and existing physics-based evaluators. Our results reveal that state-of-the-art models struggle to maintain physical validity, while MoReGen establishes a principled direction toward physically coherent video synthesis.

cs.CV↗

arXiv2024

Adaptive Finite Element Method for Phase Field Fracture Models Based on Recovery Error Estimates

The phase field model is a widely used mathematical approach for describing crack propagation in continuum damage fractures. In the context of phase field fracture simulations, adaptive finite element methods (AFEM) are often employed to address the mesh size dependency of the model. However, existing AFEM approaches for this application frequently rely on heuristic adjustments and empirical parameters for mesh refinement. In this paper, we introduce an adaptive finite element method based on a recovery type posteriori error estimates approach grounded in theoretical analysis. This method transforms the gradient of the numerical solution into a smoother function space, using the difference between the recovered gradient and the original numerical gradient as an error indicator for adaptive mesh refinement. This enables the automatic capture of crack propagation directions without the need for empirical parameters. We have implemented this adaptive method for the Hybrid formulation of the phase field model using the open-source software package FEALPy. The accuracy and efficiency of the proposed approach are demonstrated through simulations of classical 2D and 3D brittle fracture examples, validating the robustness and effectiveness of our implementation.

math.NA↗

arXiv2023

CGMI: Configurable General Multi-Agent Interaction Framework

Benefiting from the powerful capabilities of large language models (LLMs), agents based on LLMs have shown the potential to address domain-specific tasks and emulate human behaviors. However, the content generated by these agents remains somewhat superficial, owing to their limited domain expertise and the absence of an effective cognitive architecture. To address this, we present the Configurable General Multi-Agent Interaction (CGMI) framework, designed to replicate human interactions in real-world scenarios. Specifically, we propose a tree-structured methodology for the assignment, detection, and maintenance of agent personality. Additionally, we designed a cognitive architecture equipped with a skill library based on the ACT* model, which contains memory, reflection, and planning modules. We have also integrated general agents to augment the virtual environment's realism. Using the CGMI framework, we simulated numerous classroom interactions between teacher and students. The experiments indicate that aspects such as the teaching methodology, curriculum, and student performance closely mirror real classroom settings. We will open source our work.

cs.AI↗

arXiv2022

Observation of tungsten impurity suppression with ECRH by an X-ray Crystal Spectrometer on EAST

Impurity degrades tokamak plasmas confinement by causing energy loss, diluting the fuel concentration, even terminating the discharges in some extreme cases. Previously, the suppression effects of on-axis Electron Cyclotron Resonance Heating (ECRH) on the impurity accumulation have been investigated on EAST by the extreme ultraviolet (EUV) spectroscopy. However, it is difficult to quantify the changes in impurity tungsten (W) profile since the W line emissions in the EUV range could not be easily resolved. The X-ray Crystal Spectroscopy (XCS), that used to provide the ion temperature and the rotation velocity by measuring lines emissions in the soft X-ray range, also can be used to study the behavior of impurity W emissions. To begin with, in-situ absolute intensity calibration for Tangential XCS (TXCS) is conducted by analyzing the measurements of the bremsstrahlung radiation intensity. After obtaining the calibration coefficient, W44+ ion density profiles are evaluated by Abel inversion using the spectral line of W XLV (3.9095 {\AA}). Thus, a direct observation of W44+ impurity concentration suppressed by ECRH is accomplished. The obtained W density profiles can be used to analyze the W transport by combining with the impurity transport codes in the future.

physics.plasm-ph↗

arXiv2021

A long-lasting guided bone regeneration membrane from sequentially functionalised photoactive atelocollagen

The fast degradation of collagen-based membranes in the biological environment remains a critical challenge, resulting in underperforming Guided Bone Regeneration (GBR) therapy leading to compromised clinical results. Photoactive atelocollagen (AC) systems functionalised with ethylenically unsaturated monomers, such as 4-vinylbenzyl chloride (4VBC), have been shown to generate mechanically competent materials for wound healing, inflammation control and drug delivery, whereby control of the molecular architecture of the AC network is key. Building on this platform, the sequential functionalisation with 4VBC and methacrylic anhydride (MA) was hypothesised to generate UV-cured AC hydrogels with reduced swelling ratio, increased proteolytic stability and barrier functionality for GBR therapy. The sequentially functionalised atelocollagen precursor (SAP) was characterised via TNBS and ninhydrin colourimetric assays, circular dichroism and UV-curing rheometry, which confirmed nearly complete consumption of collagen primary amino groups, preserved triple helices and fast (within 180 s) gelation kinetics, respectively. Hydrogel swelling ratio and compression modulus were adjusted depending on the aqueous environment used for UV-curing, whilst the sequential functionalisation of AC successfully generated hydrogels with superior proteolytic stability in vitro compared to both 4VBC functionalised control and the commercial dental membrane Bio-Gide. These in vitro results were confirmed in vivo via both subcutaneous implantation and a proof-of-concept study in a GBR calvarial model, indicating integrity of the hydrogel and barrier defect, as well as tissue formation following 1-month implantation in rats.

q-bio.TO↗

arXiv2019

Rotation-assisted wet-spinning of UV-cured gelatin fibres and nonwovens

Photoinduced network formation is an attractive strategy for designing water-insoluble gelatin fibres as medical device building blocks and for enabling late-stage property customisation. However, mechanically-competent, long-lasting filaments are still hard to realise with current photoactive, e.g. methacrylated, gelatin systems due to inherent spinning instability and restricted coagulation capability. To explore this challenge, we present a multiscale approach combining the synthesis of 4-vinylbenzyl chloride (4VBC)-functionalised gelatin (Gel-4VBC) with a voltage-free spinning and UV-curing process so that biopolymer networks in the form of either individual fibres or nonwovens could be successfully manufactured. In comparison to state-of-the-art methacrylated gelatin, the mechanical properties of UV-cured Gel-4VBC fibres were readily modulated by adjustment of coagulation conditions, so that an ultimate tensile strength and strain at break of up to ~80 MPa and ~9 % were measured, respectively. The sequential functionalisation /spinning route proved to be highly scalable, so that one step spunlaid formation of fibroblast-friendly nonwoven fabrics was successfully demonstrated with wet spun Gel-4VBC fibres. The presented approach could be exploited to generate a library of gelatin building blocks tuneable from the molecular to the macroscopic level to deliver computer-controlled extrusion of fibres and nonwovens according to defined clinical applications.

physics.chem-ph↗

arXiv2018

Monomer-induced customisation of UV-cured atelocollagen hydrogel networks

The covalent functionalisation of type I atelocollagen with either 4-vinylbenzyl or methacrylamide residues is presented as a simple synthetic strategy to achieve customisable, cell-friendly UV-cured hydrogel networks with widespread clinical applicability. Molecular parameters, i.e. the type of monomer, degree of atelocollagen functionalisation and UV curing solution, have been systematically varied and their effect on gelation kinetics, swelling behaviour, elastic properties and enzymatic degradability investigated. UV-cured hydrogel networks deriving from atelocollagen precursors functionalised with equivalent molar content of 4-vinylbenzyl and methacrylamide adducts proved to display remarkably-different swelling ratio, storage modulus and collagenase resistance, similarly to the case of UV-cured hydrogel networks obtained with the same type of methacrylamide adduct, but varied degree of functionalisation. UV-induced network formation of 4VBC functionalised atelocollagen molecules yielded hydrogels with increased stiffness and enzymatic stability, attributed to the molecular rigidity of resulting aromatised crosslinking segment, whilst no toxic response was observed with osteosarcoma G292 cells. Although to a lesser extent, the pH of the UV-curing solution also proved to affect macroscopic hydrogel properties, likely due to the altered organisation of atelocollagen molecules during network formation. By leveraging the knowledge gained with classic synthetic networks, this study highlights how the type of monomer can be conveniently exploited to realise customisable atelocollagen hydrogels with controlled structure-property relationships to meet the requirements of unmet clinical applications.

q-bio.BM↗

arXiv2018

A UV-cured nanofibrous membrane of vinylbenzylated gelatin-poly({\epsilon}-caprolactone) dimethacrylate co-network by scalable free surface electrospinning

Electrospun nanofibrous membranes of natural polymers, such as gelatin, are fundamental in the design of regenerative devices. Crosslinking of electrospun fibres from gelatin is required to prevent dissolution in water, to retain the original nanofibre morphology after immersion in water, and to improve the thermal and mechanical properties, although this is still challenging to accomplish in a controlled fashion. In this study, we have investigated the scalable manufacture and structural stability in aqueous environment of a UV-cured nanofibrous membrane fabricated by free surface electrospinning (FSES) of aqueous solutions containing vinylbenzylated gelatin and poly(epsilon-caprolactone) dimethacrylate (PCL-DMA). Vinylbenzylated gelatin was obtained via chemical functionalisation with photopolymerisable 4-vinylbenzyl chloride (4VBC) groups, so that the gelatin and PCL phase in electrospun fibres were integrated in a covalent UV-cured co-network at the molecular scale, rather than being simply physically mixed. UV-cured nanofibrous membranes did not dissolve in water and showed enhanced thermal and mechanical properties, with respect to as-spun samples, indicating the effectiveness of the photo-crosslinking reaction. In addition, UV-cured gelatin/PCL membranes displayed increased structural stability in water with respect to PCL-free samples and were highly tolerated by G292 osteosarcoma cells. These results therefore support the use of PCL DMA as hydrophobic, biodegradable crosslinker and provide new insight on the scalable design of water insoluble, mechanical competent gelatin membranes for healthcare applications.

physics.app-ph↗

arXiv2016

Protease-sensitive atelocollagen hydrogels promote healing in a diabetic wound model

The design of exudate-managing wound dressings is an established route to accelerated healing, although such design remains a challenge from material and manufacturing standpoints. Aiming towards the clinical translation of knowledge gained in vitro with highly swollen rat tail collagen hydrogels, this study investigated the healing capability in a diabetic mouse wound model of telopeptide-free, protease-inhibiting collagen networks. 4 vinylbenzylation and UV irradiation of type I atelocollagen (AC) led to hydrogel networks with chemical and macroscopic properties comparable to previous collagen analogues, attributable to similar lysine content and dichroic properties. After 4 days in vitro, hydrogels induced nearly 50 RFU% reduction in matrix metalloproteinase (MMP)-9 activity, whilst showing less than 20 wt.-% weight loss. After 20 days in vivo, dry networks promoted 99% closure of 10x10 mm full thickness wounds and accelerated neodermal tissue formation compared to Mepilex. This collagen system can be equipped with multiple, customisable properties and functions key to personalised chronic wound care.

physics.bio-ph↗