Selected work

Representative Papers

Rectifying Geometric Misalignment: Online Source-Free Adaptation for Class-Imbalanced EEG

Aug 05, 2026

This work addresses the degradation of Riemannian manifold geometric alignment in online brain–computer interfaces caused by label shift. To tackle this issue without access to source-domain data, the authors propose the Online Streaming Plug-and-Play Domain-Invariant Manifold (OSPDIM) framework, which introduces manifold-constrained bias into tangent space mapping for the first time. By integrating an online information maximization criterion, OSPDIM adaptively optimizes bias parameters in real time, enabling plug-and-play geometric correction without reliance on historical batch statistics. Experimental results demonstrate that OSPDIM significantly outperforms standard Riemannian methods across multiple motor imagery EEG datasets, exhibiting exceptional robustness—particularly under severe class imbalance in online settings.

0 citationsRead paper

SHAPE: Simultaneous Water Hydraulic Actuation and Position Estimation of a Sensorless Remote Actuator through a Thin and Long Flexible Tube

Jul 18, 2026

This work addresses the challenge of precise position control in long, flexible, low-impedance hydraulic actuation systems operating in harsh environments, where conventional robots suffer from sensor fragility and the impracticality of embedding sensors. The authors propose an innovative approach that simultaneously transmits actuation power and state information through a single water-filled flexible tube. By modeling volume loss due to pressure-induced tube deformation and the subtle effects of entrained air, they achieve high-accuracy, sensorless position estimation and actuation over tube lengths up to 50 meters. Key contributions include the first demonstration of integrated drive and position feedback using only a single hydraulic line without end-effector sensors, and a practical online parameter identification method to compensate for inter-tube variability and air content fluctuations. Experiments validate stable position control of hydraulic cylinders under varying loads, establishing a robust framework for remote robotic operation in extreme conditions.

0 citationsRead paper

Actuator Reality Shaping for Zero-Shot Sim-to-Real Robot Learning

Jul 02, 2026

This work addresses the sim-to-real transfer failure commonly encountered in reinforcement learning due to mismatches between idealized actuator models used in simulation and the nonlinear, hardware-dependent motor dynamics of real robots. To bridge this gap, the authors propose “actuator reality shaping,” a method that deploys a two-degree-of-freedom feedforward–feedback controller on physical hardware to shape the closed-loop actuator response to closely match an ideal second-order reference model assumed in simulation. Notably, this approach requires no system identification or learned actuator models and enables zero-shot policy deployment through a standardized actuator interface. Experiments across diverse platforms—including single-joint servos, a 7-DoF manipulator, wheeled-legged robots, and humanoids—demonstrate substantial reductions in tracking error and successful zero-shot transfer across multiple tasks and systems, thereby shifting the paradigm from increasing simulation fidelity to unifying real-world actuator behavior to conform to simulation assumptions.

0 citationsRead paper

Regularized Reward-Punishment Reinforcement Learning

Jun 26, 2026

This work addresses the challenge in traditional reinforcement learning where reward and punishment mechanisms are typically optimized independently, hindering effective coordination. To overcome this limitation, the authors propose the KL-Coupled Policy Regularization (KCPR) framework, which enables direct interaction between reward and punishment signals at the policy level for the first time. The approach employs mutual companion policies as soft priors, jointly optimizing policy updates and value propagation. It introduces a KL-regularized Bellman operator, the klDMP deep learning algorithm, and a dual replay buffer mechanism. Evaluated on grid-world environments and Gazebo-based robotic navigation tasks, klDMP demonstrates significantly improved learning stability and safety while maintaining task performance comparable to established methods such as DQN, SQL, and softDMP.

0 citationsRead paper

Self Capacitive Tactile Sensor System designed for Companion Robots

Jun 23, 2026

This study addresses the challenges of complexity, wiring overhead, and high cost in existing tactile sensing systems, which hinder low-latency, full-body haptic perception in companion robots. The authors propose a streamlined and scalable architecture based on self-capacitance principles, requiring only a single layer of conductive fabric and conductive threads—eliminating the need for intricate electrode patterning. By integrating flexible printed circuits with an FPGA-based embedded platform, they deploy a lightweight decision tree classifier to enable low-power, low-latency edge inference. The resulting system successfully implements a 100-sensor-point flexible array capable of accurately distinguishing four interaction types—light touch, slow tap, fast tap, and strike—at sampling rates of at least 100 Hz, thereby fulfilling the whole-body tactile sensing requirements of the HIRO-chan companion robot.

0 citationsRead paper

Recent publications

Latest Papers

Rectifying Geometric Misalignment: Online Source-Free Adaptation for Class-Imbalanced EEG

Aug 05, 2026

This work addresses the degradation of Riemannian manifold geometric alignment in online brain–computer interfaces caused by label shift. To tackle this issue without access to source-domain data, the authors propose the Online Streaming Plug-and-Play Domain-Invariant Manifold (OSPDIM) framework, which introduces manifold-constrained bias into tangent space mapping for the first time. By integrating an online information maximization criterion, OSPDIM adaptively optimizes bias parameters in real time, enabling plug-and-play geometric correction without reliance on historical batch statistics. Experimental results demonstrate that OSPDIM significantly outperforms standard Riemannian methods across multiple motor imagery EEG datasets, exhibiting exceptional robustness—particularly under severe class imbalance in online settings.

0 citationsRead paper

SHAPE: Simultaneous Water Hydraulic Actuation and Position Estimation of a Sensorless Remote Actuator through a Thin and Long Flexible Tube

Jul 18, 2026

This work addresses the challenge of precise position control in long, flexible, low-impedance hydraulic actuation systems operating in harsh environments, where conventional robots suffer from sensor fragility and the impracticality of embedding sensors. The authors propose an innovative approach that simultaneously transmits actuation power and state information through a single water-filled flexible tube. By modeling volume loss due to pressure-induced tube deformation and the subtle effects of entrained air, they achieve high-accuracy, sensorless position estimation and actuation over tube lengths up to 50 meters. Key contributions include the first demonstration of integrated drive and position feedback using only a single hydraulic line without end-effector sensors, and a practical online parameter identification method to compensate for inter-tube variability and air content fluctuations. Experiments validate stable position control of hydraulic cylinders under varying loads, establishing a robust framework for remote robotic operation in extreme conditions.

0 citationsRead paper

Actuator Reality Shaping for Zero-Shot Sim-to-Real Robot Learning

Jul 02, 2026

This work addresses the sim-to-real transfer failure commonly encountered in reinforcement learning due to mismatches between idealized actuator models used in simulation and the nonlinear, hardware-dependent motor dynamics of real robots. To bridge this gap, the authors propose “actuator reality shaping,” a method that deploys a two-degree-of-freedom feedforward–feedback controller on physical hardware to shape the closed-loop actuator response to closely match an ideal second-order reference model assumed in simulation. Notably, this approach requires no system identification or learned actuator models and enables zero-shot policy deployment through a standardized actuator interface. Experiments across diverse platforms—including single-joint servos, a 7-DoF manipulator, wheeled-legged robots, and humanoids—demonstrate substantial reductions in tracking error and successful zero-shot transfer across multiple tasks and systems, thereby shifting the paradigm from increasing simulation fidelity to unifying real-world actuator behavior to conform to simulation assumptions.

0 citationsRead paper

Regularized Reward-Punishment Reinforcement Learning

Jun 26, 2026

This work addresses the challenge in traditional reinforcement learning where reward and punishment mechanisms are typically optimized independently, hindering effective coordination. To overcome this limitation, the authors propose the KL-Coupled Policy Regularization (KCPR) framework, which enables direct interaction between reward and punishment signals at the policy level for the first time. The approach employs mutual companion policies as soft priors, jointly optimizing policy updates and value propagation. It introduces a KL-regularized Bellman operator, the klDMP deep learning algorithm, and a dual replay buffer mechanism. Evaluated on grid-world environments and Gazebo-based robotic navigation tasks, klDMP demonstrates significantly improved learning stability and safety while maintaining task performance comparable to established methods such as DQN, SQL, and softDMP.

0 citationsRead paper

Self Capacitive Tactile Sensor System designed for Companion Robots

Jun 23, 2026

This study addresses the challenges of complexity, wiring overhead, and high cost in existing tactile sensing systems, which hinder low-latency, full-body haptic perception in companion robots. The authors propose a streamlined and scalable architecture based on self-capacitance principles, requiring only a single layer of conductive fabric and conductive threads—eliminating the need for intricate electrode patterning. By integrating flexible printed circuits with an FPGA-based embedded platform, they deploy a lightweight decision tree classifier to enable low-power, low-latency edge inference. The resulting system successfully implements a 100-sensor-point flexible array capable of accurately distinguishing four interaction types—light touch, slow tap, fast tap, and strike—at sampling rates of at least 100 Hz, thereby fulfilling the whole-body tactile sensing requirements of the HIRO-chan companion robot.

0 citationsRead paper