Selected work

Representative Papers

Hybrid Framework for Robotic Manipulation: Integrating Reinforcement Learning and Large Language Models

Mar 31, 2026

This work addresses the challenge of enabling robots to interpret complex natural language instructions and execute low-level actions efficiently in dynamic environments. To bridge the gap between high-level reasoning and low-level control, we propose a novel framework that deeply integrates large language models (LLMs) with reinforcement learning (RL): the LLM handles high-level task planning and semantic understanding, while RL governs precise low-level motor control. The system is evaluated in both PyBullet simulation and on a physical Franka Emika Panda robotic arm. Compared to pure RL baselines, our approach reduces task completion time by 33.5%, improves execution accuracy by 18.1%, and enhances environmental adaptability by 36.4%, demonstrating real-time, natural language–driven adaptive robot manipulation.

0 citationsRead paper

Explainable Knowledge Distillation for Efficient Medical Image Classification

Aug 21, 2025

To address the dual challenges of efficiency and interpretability in diagnosing COVID-19 and lung cancer from chest X-rays (CXRs) under resource-constrained clinical settings, this work proposes a hybrid-supervised knowledge distillation framework. Methodologically, it employs high-capacity teacher models—VGG19, Visformer-S, and AutoFormer-V2-T—and constructs a hardware-aware lightweight student model based on the OFA-595 supernet. Training integrates joint optimization with both soft and ground-truth labels, while Score-CAM enables spatially interpretable decision visualization. Experiments on the COVID-QU-Ex and LCS25000 datasets demonstrate that the student model reduces parameter count by over 80%, achieves 3.2× inference speedup, and attains classification accuracy only 1.3% lower than the teachers. Moreover, it exhibits strong localization consistency and clinically trustworthy interpretability. These advances significantly enhance the deployability of lightweight models in primary healthcare settings.

0 citationsRead paper

Geometric Determinations Of Characteristic Redshifts From DESI-DR2 BAO and DES-SN5YR Observations: Hints For New Expansion Rate Anomalies

May 25, 2025

A significant tension persists between late-time cosmic expansion measurements and the Planck 2018 ΛCDM prediction. Method: We perform a model-independent cosmological reconstruction using DESI-DR2 baryon acoustic oscillation (BAO) and DES-SN5YR supernova data, jointly applying Gaussian process regression and node spline interpolation to estimate the distance modulus and its derivatives. Contribution/Results: For the first time, we robustly identify a 4–5σ anomaly in the Hubble parameter *H(z)* within *z* ∼ 0.35–0.55, precisely localizing this feature as a sensitive probe of new physics. Our nonparametric approach avoids assumptions about specific cosmological models, substantially improving constraints on the dimensionless Hubble function *E(z)*. Consistent deviations from ΛCDM are confirmed in both BAO-only and BAO+SN joint analyses, providing critical geometric evidence for evolving dark energy or modified gravity theories.

0 citationsRead paper

A Robust and Energy-Efficient Trajectory Planning Framework for High-Degree-of-Freedom Robots

Mar 13, 2025

To address the coupled optimization challenge of high energy consumption and motion non-smoothness in high-degree-of-freedom robot trajectory planning, this paper proposes a synergistic optimization framework integrating sinusoidal trajectory parameterization with dynamic velocity scaling. Leveraging physics-based simulation modeling, joint trajectories are explicitly represented using sine functions, while a real-time adaptive velocity scaling strategy jointly optimizes energy consumption and motion smoothness—achieving sub-millimeter accuracy (position error ≤ 0.3 mm). Simulation results demonstrate a 32% reduction in energy consumption and a 41% decrease in trajectory jitter compared to baseline methods, significantly enhancing energy efficiency, robustness, and mechanical longevity. This work constitutes the first systematic integration of sinusoidal parameterization and dynamic scaling for high-dimensional robotic motion planning, establishing a novel, interpretable, and deployable paradigm for energy-efficient, smooth trajectory generation.

0 citationsRead paper

A New $sim 5sigma$ Tension at Characteristic Redshift from DESI DR1 and DES-SN5YR observations

Mar 04, 2025

This work addresses the >5σ tension between low-redshift (z ∼ 0.512) H(z) measurements and Planck 2018 ΛCDM predictions—a critical challenge to standard cosmology. We propose a model-independent multi-task Gaussian process (MTGP) method that jointly reconstructs the angular diameter distance D_A from DESI DR1 baryon acoustic oscillation (BAO) and DES-SN5YR supernova data, then derives H(z) using the Planck-calibrated sound horizon scale r_d. Our reconstruction yields the first >5σ deviation from ΛCDM at z ∼ 0.512, while showing full consistency with the model at z ∼ 1.63—strongly disfavoring dominant systematic errors. Unlike the well-known H₀ tension, this low-redshift H(z) anomaly may constitute the earliest robust evidence for new physics, such as evolving dark energy or modified gravity, thereby opening a novel observational dimension for testing the ΛCDM paradigm.

0 citationsRead paper

Recent publications

Latest Papers

Hybrid Framework for Robotic Manipulation: Integrating Reinforcement Learning and Large Language Models

Mar 31, 2026

This work addresses the challenge of enabling robots to interpret complex natural language instructions and execute low-level actions efficiently in dynamic environments. To bridge the gap between high-level reasoning and low-level control, we propose a novel framework that deeply integrates large language models (LLMs) with reinforcement learning (RL): the LLM handles high-level task planning and semantic understanding, while RL governs precise low-level motor control. The system is evaluated in both PyBullet simulation and on a physical Franka Emika Panda robotic arm. Compared to pure RL baselines, our approach reduces task completion time by 33.5%, improves execution accuracy by 18.1%, and enhances environmental adaptability by 36.4%, demonstrating real-time, natural language–driven adaptive robot manipulation.

0 citationsRead paper

Explainable Knowledge Distillation for Efficient Medical Image Classification

Aug 21, 2025

To address the dual challenges of efficiency and interpretability in diagnosing COVID-19 and lung cancer from chest X-rays (CXRs) under resource-constrained clinical settings, this work proposes a hybrid-supervised knowledge distillation framework. Methodologically, it employs high-capacity teacher models—VGG19, Visformer-S, and AutoFormer-V2-T—and constructs a hardware-aware lightweight student model based on the OFA-595 supernet. Training integrates joint optimization with both soft and ground-truth labels, while Score-CAM enables spatially interpretable decision visualization. Experiments on the COVID-QU-Ex and LCS25000 datasets demonstrate that the student model reduces parameter count by over 80%, achieves 3.2× inference speedup, and attains classification accuracy only 1.3% lower than the teachers. Moreover, it exhibits strong localization consistency and clinically trustworthy interpretability. These advances significantly enhance the deployability of lightweight models in primary healthcare settings.

0 citationsRead paper

Geometric Determinations Of Characteristic Redshifts From DESI-DR2 BAO and DES-SN5YR Observations: Hints For New Expansion Rate Anomalies

May 25, 2025

A significant tension persists between late-time cosmic expansion measurements and the Planck 2018 ΛCDM prediction. Method: We perform a model-independent cosmological reconstruction using DESI-DR2 baryon acoustic oscillation (BAO) and DES-SN5YR supernova data, jointly applying Gaussian process regression and node spline interpolation to estimate the distance modulus and its derivatives. Contribution/Results: For the first time, we robustly identify a 4–5σ anomaly in the Hubble parameter *H(z)* within *z* ∼ 0.35–0.55, precisely localizing this feature as a sensitive probe of new physics. Our nonparametric approach avoids assumptions about specific cosmological models, substantially improving constraints on the dimensionless Hubble function *E(z)*. Consistent deviations from ΛCDM are confirmed in both BAO-only and BAO+SN joint analyses, providing critical geometric evidence for evolving dark energy or modified gravity theories.

0 citationsRead paper

A Robust and Energy-Efficient Trajectory Planning Framework for High-Degree-of-Freedom Robots

Mar 13, 2025

To address the coupled optimization challenge of high energy consumption and motion non-smoothness in high-degree-of-freedom robot trajectory planning, this paper proposes a synergistic optimization framework integrating sinusoidal trajectory parameterization with dynamic velocity scaling. Leveraging physics-based simulation modeling, joint trajectories are explicitly represented using sine functions, while a real-time adaptive velocity scaling strategy jointly optimizes energy consumption and motion smoothness—achieving sub-millimeter accuracy (position error ≤ 0.3 mm). Simulation results demonstrate a 32% reduction in energy consumption and a 41% decrease in trajectory jitter compared to baseline methods, significantly enhancing energy efficiency, robustness, and mechanical longevity. This work constitutes the first systematic integration of sinusoidal parameterization and dynamic scaling for high-dimensional robotic motion planning, establishing a novel, interpretable, and deployable paradigm for energy-efficient, smooth trajectory generation.

0 citationsRead paper

A New $sim 5sigma$ Tension at Characteristic Redshift from DESI DR1 and DES-SN5YR observations

Mar 04, 2025

This work addresses the >5σ tension between low-redshift (z ∼ 0.512) H(z) measurements and Planck 2018 ΛCDM predictions—a critical challenge to standard cosmology. We propose a model-independent multi-task Gaussian process (MTGP) method that jointly reconstructs the angular diameter distance D_A from DESI DR1 baryon acoustic oscillation (BAO) and DES-SN5YR supernova data, then derives H(z) using the Planck-calibrated sound horizon scale r_d. Our reconstruction yields the first >5σ deviation from ΛCDM at z ∼ 0.512, while showing full consistency with the model at z ∼ 1.63—strongly disfavoring dominant systematic errors. Unlike the well-known H₀ tension, this low-redshift H(z) anomaly may constitute the earliest robust evidence for new physics, such as evolving dark energy or modified gravity, thereby opening a novel observational dimension for testing the ΛCDM paradigm.

0 citationsRead paper