AGRO-Nav: Autonomous Graph-based Orchard Navigation
AGRO-Nav通过构建基于SLAM点云的拓扑图并使用Dijkstra搜索和Theta*算法，解决了果园中自动导航偏离行中心及碰撞风险的问题。
0 citationsRead paper
Selected work
Representative Papers
CORAL: Curriculum-Optimized Reward Adaptation for LiDAR-Based Goal-Directed Urban Driving
This study addresses the challenge of协同 learning for multi-objective navigation strategies in urban autonomous driving by proposing the CORAL framework. This approach introduces a novel joint optimization mechanism integrating curriculum learning with dynamic rewards, leveraging polar LiDAR histograms and the PPO algorithm to efficiently learn complex behavioral constraints without relying on point cloud encoders or BEV representations. Experimental results demonstrate that CORAL achieves a 100% success rate on the longest routes and exhibits strong zero-shot transferability to seven new towns with success rates ranging from 68% to 98%, while maintaining lateral deviations below 0.35 m. Consequently, this framework enables efficient training and robust generalization of long-horizon goal-oriented driving policies under complex constraints.
0 citationsRead paper
UART for Wearables (U4We): DC Power and Carrierless Signal Transfer over Conductive Textiles
This work addresses the lack of efficient, wiring-free co-transmission mechanisms for power and data among batteryless distributed wearable modules by proposing the first carrier-free fabric-based co-transmission architecture leveraging UART signaling. The approach employs a transmission line formed by two conductive fabric layers separated by an insulating layer, enabling direct AC-coupled injection of UART pulses to simultaneously deliver DC power and data without carrier modulation. Signal recovery is achieved via a comparator informed by a second-order circuit transient response model, and an analytical framework is established to assess the feasibility of data rates, fabric parameters, and decoupling inductor design. Experimental results demonstrate successful high-bit-rate synchronous transmission of both power and data over conductive fabrics, offering a low-overhead interconnect foundation for wearable systems.
0 citationsRead paper
Transferable Low-Rank Convolutional Bases for Onboarding Unseen Medical Imaging Modalities
This work addresses the challenge of efficiently adapting medical imaging models to unseen modalities post-deployment while avoiding catastrophic forgetting. The authors propose a parameter-efficient adaptation method that, under a strict leave-one-domain-out setting, freezes the pre-trained convolutional backbone and leverages transferable low-rank convolutional bases learned from source modalities. Adaptation is achieved solely through projection parameters atop these bases, constituting only 0.78% of the total model parameters. By integrating convolutional LoRA, low-rank decomposition, and Mahalanobis anomaly detection, the approach improves adaptation accuracy on new modalities by 6.11 percentage points over random bases, while incurring zero performance loss on source modalities (Δ = 0.00 pp), substantially outperforming full fine-tuning and decision-level adaptation strategies.
0 citationsRead paper
Morphological Analysis of Semiconductor Microstructures using Skeleton Graphs
This study investigates the relative influence of ion beam irradiation angle versus fluence on microstructural morphology evolution of germanium (Ge) surfaces. We propose a quantitative analytical framework integrating skeleton-graph topological characterization with graph convolutional network (GCN) embedding, coupled with principal component analysis (PCA) and the Davies–Bouldin index to evaluate class separability of microstructures across irradiation conditions. Results demonstrate that irradiation angle is the dominant parameter governing surface morphological evolution—its effect substantially outweighs that of fluence. The framework enables unsupervised, interpretable discrimination of microstructural patterns. To our knowledge, this work represents the first application of graph neural networks to ion-beam irradiation-induced morphology analysis. It establishes a novel paradigm for precise microstructural engineering of semiconductor materials and provides a generalizable computational toolkit for quantitative microstructure–processing–property analysis.
0 citationsRead paper
Recent publications
Latest Papers
AGRO-Nav: Autonomous Graph-based Orchard Navigation
AGRO-Nav通过构建基于SLAM点云的拓扑图并使用Dijkstra搜索和Theta*算法，解决了果园中自动导航偏离行中心及碰撞风险的问题。
0 citationsRead paper
CORAL: Curriculum-Optimized Reward Adaptation for LiDAR-Based Goal-Directed Urban Driving
This study addresses the challenge of协同 learning for multi-objective navigation strategies in urban autonomous driving by proposing the CORAL framework. This approach introduces a novel joint optimization mechanism integrating curriculum learning with dynamic rewards, leveraging polar LiDAR histograms and the PPO algorithm to efficiently learn complex behavioral constraints without relying on point cloud encoders or BEV representations. Experimental results demonstrate that CORAL achieves a 100% success rate on the longest routes and exhibits strong zero-shot transferability to seven new towns with success rates ranging from 68% to 98%, while maintaining lateral deviations below 0.35 m. Consequently, this framework enables efficient training and robust generalization of long-horizon goal-oriented driving policies under complex constraints.
0 citationsRead paper
UART for Wearables (U4We): DC Power and Carrierless Signal Transfer over Conductive Textiles
This work addresses the lack of efficient, wiring-free co-transmission mechanisms for power and data among batteryless distributed wearable modules by proposing the first carrier-free fabric-based co-transmission architecture leveraging UART signaling. The approach employs a transmission line formed by two conductive fabric layers separated by an insulating layer, enabling direct AC-coupled injection of UART pulses to simultaneously deliver DC power and data without carrier modulation. Signal recovery is achieved via a comparator informed by a second-order circuit transient response model, and an analytical framework is established to assess the feasibility of data rates, fabric parameters, and decoupling inductor design. Experimental results demonstrate successful high-bit-rate synchronous transmission of both power and data over conductive fabrics, offering a low-overhead interconnect foundation for wearable systems.
0 citationsRead paper
Transferable Low-Rank Convolutional Bases for Onboarding Unseen Medical Imaging Modalities
This work addresses the challenge of efficiently adapting medical imaging models to unseen modalities post-deployment while avoiding catastrophic forgetting. The authors propose a parameter-efficient adaptation method that, under a strict leave-one-domain-out setting, freezes the pre-trained convolutional backbone and leverages transferable low-rank convolutional bases learned from source modalities. Adaptation is achieved solely through projection parameters atop these bases, constituting only 0.78% of the total model parameters. By integrating convolutional LoRA, low-rank decomposition, and Mahalanobis anomaly detection, the approach improves adaptation accuracy on new modalities by 6.11 percentage points over random bases, while incurring zero performance loss on source modalities (Δ = 0.00 pp), substantially outperforming full fine-tuning and decision-level adaptation strategies.
0 citationsRead paper
Morphological Analysis of Semiconductor Microstructures using Skeleton Graphs
This study investigates the relative influence of ion beam irradiation angle versus fluence on microstructural morphology evolution of germanium (Ge) surfaces. We propose a quantitative analytical framework integrating skeleton-graph topological characterization with graph convolutional network (GCN) embedding, coupled with principal component analysis (PCA) and the Davies–Bouldin index to evaluate class separability of microstructures across irradiation conditions. Results demonstrate that irradiation angle is the dominant parameter governing surface morphological evolution—its effect substantially outweighs that of fluence. The framework enables unsupervised, interpretable discrimination of microstructural patterns. To our knowledge, this work represents the first application of graph neural networks to ion-beam irradiation-induced morphology analysis. It establishes a novel paradigm for precise microstructural engineering of semiconductor materials and provides a generalizable computational toolkit for quantitative microstructure–processing–property analysis.
0 citationsRead paper