Selected work

Representative Papers

Robot Planning and Situation Handling with Active Perception

Apr 28, 2026

This work addresses the challenge of task execution failures in dynamic, open-world environments—such as those caused by jammed doors or unforeseen ground obstacles—by introducing the VAP-TAMP framework. VAP-TAMP uniquely integrates action-knowledge-guided active viewpoint selection with vision-language models and leverages scene graph construction and reasoning to enable joint task and motion planning (TAMP). The proposed approach facilitates real-time detection of and response to execution anomalies, significantly improving both task success rates and robotic autonomy in complex, dynamic settings. Evaluations on both simulated and real-world service robot platforms demonstrate its effectiveness in enhancing robustness and adaptability under uncertainty.

0 citationsRead paper

Optimal Labeler Assignment and Sampling for Active Learning in the Presence of Imperfect Labels

Dec 14, 2025

To address high label noise in active learning caused by annotator ability disparities—particularly erroneous labeling of complex instances—this paper proposes a robust, noise-resilient active learning framework. Methodologically: (1) it formulates an optimal annotator allocation model grounded in game theory, minimizing the worst-case potential noise per iteration; (2) it introduces an uncertainty-aware, noise-robust sampling strategy; and (3) it integrates multi-annotator confidence-weighted ensemble learning with noise-robust loss modeling. Extensive experiments across multiple benchmark datasets demonstrate an average 5.2% improvement in classification accuracy and a 37% reduction in label-noise sensitivity, significantly outperforming state-of-the-art active learning methods. The core contribution lies in the first unified integration of annotator capability modeling, noise-aware sampling, and robust ensemble learning within a closed-loop active learning pipeline.

0 citationsRead paper

GOMP: Grasped Object Manifold Projection for Multimodal Imitation Learning of Manipulation

Dec 03, 2025

To address trajectory inaccuracy in imitation learning caused by error accumulation during precision assembly, this paper proposes a novel imitation learning framework constrained by a low-dimensional manifold of the manipulated object. The core innovation is the first introduction of a manifold projection mechanism that constrains object motion to a task-relevant low-dimensional manifold, coupled with an *n*-armed bandit algorithm for dynamic policy adaptation—effectively suppressing error propagation without requiring additional labeled data. The method integrates multimodal perception (including tactile sensing), non-rigid object modeling, and manifold-aware optimization, yielding strong generalization capability. Evaluated on four high-precision assembly tasks, the approach achieves significant improvements in both success rate and trajectory accuracy, demonstrating its cross-modal effectiveness and engineering practicality.

0 citationsRead paper

BudgetMem: Learning Selective Memory Policies for Cost-Efficient Long-Context Processing in Language Models

Nov 07, 2025

To address the prohibitively high computational and memory overhead of large language models (LLMs) in long-context processing, this paper proposes a budget-constrained Selective Memory architecture. The method innovatively integrates a learnable memory gating mechanism with a multidimensional importance scoring function—incorporating BM25, entity density, TF-IDF, discourse markers, and positional bias—to dynamically select and retain salient information under strict memory constraints, thereby departing from the conventional full-document retrieval-augmented generation (RAG) paradigm. Evaluated on an enhanced Llama-3.2-3B-Instruct model, the approach achieves only a 1.0% F1 drop on long-document tasks while reducing memory consumption by 72.4%. Crucially, its performance advantage over baseline RAG methods grows substantially with increasing document length, demonstrating superior scalability and efficiency in resource-constrained long-context settings.

0 citationsRead paper

Differentiable Constraint-Based Causal Discovery

Oct 24, 2025

Causal discovery aims to infer causal structures among variables from observational data, forming the foundation for AI-driven decision-making and intervention. Existing constraint-based methods (e.g., PC) rely on conditional independence tests but suffer from poor statistical power in small-sample regimes; score-based approaches (e.g., NOTEARS) enable differentiable optimization yet neglect explicit independence constraints. This paper proposes the first differentiable d-separation scoring framework integrating soft logic and percolation theory: conditional independencies are encoded as soft logical formulas; a continuous, differentiable d-separation metric is constructed via percolation-theoretic principles; and the resulting objective is embedded into a gradient-based end-to-end learning framework for causal graph estimation. Our method unifies the statistical rigor of constraint-based methods with the optimization flexibility of score-based ones. It achieves significant improvements over state-of-the-art baselines under limited samples and establishes new SOTA performance across multiple real-world benchmarks. Code and data are publicly available.

0 citationsRead paper

Recent publications

Latest Papers

Robot Planning and Situation Handling with Active Perception

Apr 28, 2026

This work addresses the challenge of task execution failures in dynamic, open-world environments—such as those caused by jammed doors or unforeseen ground obstacles—by introducing the VAP-TAMP framework. VAP-TAMP uniquely integrates action-knowledge-guided active viewpoint selection with vision-language models and leverages scene graph construction and reasoning to enable joint task and motion planning (TAMP). The proposed approach facilitates real-time detection of and response to execution anomalies, significantly improving both task success rates and robotic autonomy in complex, dynamic settings. Evaluations on both simulated and real-world service robot platforms demonstrate its effectiveness in enhancing robustness and adaptability under uncertainty.

0 citationsRead paper

Optimal Labeler Assignment and Sampling for Active Learning in the Presence of Imperfect Labels

Dec 14, 2025

To address high label noise in active learning caused by annotator ability disparities—particularly erroneous labeling of complex instances—this paper proposes a robust, noise-resilient active learning framework. Methodologically: (1) it formulates an optimal annotator allocation model grounded in game theory, minimizing the worst-case potential noise per iteration; (2) it introduces an uncertainty-aware, noise-robust sampling strategy; and (3) it integrates multi-annotator confidence-weighted ensemble learning with noise-robust loss modeling. Extensive experiments across multiple benchmark datasets demonstrate an average 5.2% improvement in classification accuracy and a 37% reduction in label-noise sensitivity, significantly outperforming state-of-the-art active learning methods. The core contribution lies in the first unified integration of annotator capability modeling, noise-aware sampling, and robust ensemble learning within a closed-loop active learning pipeline.

0 citationsRead paper

GOMP: Grasped Object Manifold Projection for Multimodal Imitation Learning of Manipulation

Dec 03, 2025

To address trajectory inaccuracy in imitation learning caused by error accumulation during precision assembly, this paper proposes a novel imitation learning framework constrained by a low-dimensional manifold of the manipulated object. The core innovation is the first introduction of a manifold projection mechanism that constrains object motion to a task-relevant low-dimensional manifold, coupled with an *n*-armed bandit algorithm for dynamic policy adaptation—effectively suppressing error propagation without requiring additional labeled data. The method integrates multimodal perception (including tactile sensing), non-rigid object modeling, and manifold-aware optimization, yielding strong generalization capability. Evaluated on four high-precision assembly tasks, the approach achieves significant improvements in both success rate and trajectory accuracy, demonstrating its cross-modal effectiveness and engineering practicality.

0 citationsRead paper

BudgetMem: Learning Selective Memory Policies for Cost-Efficient Long-Context Processing in Language Models

Nov 07, 2025

To address the prohibitively high computational and memory overhead of large language models (LLMs) in long-context processing, this paper proposes a budget-constrained Selective Memory architecture. The method innovatively integrates a learnable memory gating mechanism with a multidimensional importance scoring function—incorporating BM25, entity density, TF-IDF, discourse markers, and positional bias—to dynamically select and retain salient information under strict memory constraints, thereby departing from the conventional full-document retrieval-augmented generation (RAG) paradigm. Evaluated on an enhanced Llama-3.2-3B-Instruct model, the approach achieves only a 1.0% F1 drop on long-document tasks while reducing memory consumption by 72.4%. Crucially, its performance advantage over baseline RAG methods grows substantially with increasing document length, demonstrating superior scalability and efficiency in resource-constrained long-context settings.

0 citationsRead paper

Differentiable Constraint-Based Causal Discovery

Oct 24, 2025

Causal discovery aims to infer causal structures among variables from observational data, forming the foundation for AI-driven decision-making and intervention. Existing constraint-based methods (e.g., PC) rely on conditional independence tests but suffer from poor statistical power in small-sample regimes; score-based approaches (e.g., NOTEARS) enable differentiable optimization yet neglect explicit independence constraints. This paper proposes the first differentiable d-separation scoring framework integrating soft logic and percolation theory: conditional independencies are encoded as soft logical formulas; a continuous, differentiable d-separation metric is constructed via percolation-theoretic principles; and the resulting objective is embedded into a gradient-based end-to-end learning framework for causal graph estimation. Our method unifies the statistical rigor of constraint-based methods with the optimization flexibility of score-based ones. It achieves significant improvements over state-of-the-art baselines under limited samples and establishes new SOTA performance across multiple real-world benchmarks. Code and data are publicly available.

0 citationsRead paper