Selected work

Representative Papers

Hybrid Data-Driven Predictive Control for Robust and Reactive Exoskeleton Locomotion Synthesis

Aug 13, 2025

To address the challenge of achieving robust, reactive bipedal locomotion for exoskeleton robots in dynamic environments, this paper proposes a hybrid data-driven predictive control framework. The method innovatively embeds inter-step transition modeling into model predictive control (MPC), enabling unified optimization of discrete contact sequences and continuous motion trajectories while supporting online replanning. By representing system dynamics via Hankel matrices, it integrates data-driven predictive control with a step-to-step (S2S) state transition model, synthesizing motion and responding to real-time disturbances using only historical input–output data—without requiring explicit dynamical models. Experimental validation on the Atalante exoskeleton platform demonstrates significant improvements in walking robustness and environmental adaptability, enabling stable bipedal gait under complex, time-varying conditions.

0 citationsRead paper

Leg Exoskeleton Odometry using a Limited FOV Depth Sensor

Feb 26, 2025

In lower-limb exoskeletons, narrow field-of-view (FoV) depth sensors—constrained by mechanical mounting—suffer severe motion-induced distortion, leading to significant odometry drift. To address this, we propose an EKF-ICP collaborative fusion framework that tightly integrates narrow-FoV depth point clouds with exoskeleton proprioceptive measurements (joint encoders and IMU). State estimation is performed via an extended Kalman filter (EKF), while a customized iterative closest point (ICP) algorithm enhances point cloud registration robustness under strong motion disturbances, enabling high-fidelity terrain elevation mapping. Experimental results demonstrate a 62% reduction in horizontal pose drift compared to a pure proprioceptive baseline; moreover, the generated elevation maps exhibit substantially improved completeness and geometric consistency over conventional point cloud mapping approaches.

0 citationsRead paper

Recent publications

Latest Papers

Hybrid Data-Driven Predictive Control for Robust and Reactive Exoskeleton Locomotion Synthesis

Aug 13, 2025

To address the challenge of achieving robust, reactive bipedal locomotion for exoskeleton robots in dynamic environments, this paper proposes a hybrid data-driven predictive control framework. The method innovatively embeds inter-step transition modeling into model predictive control (MPC), enabling unified optimization of discrete contact sequences and continuous motion trajectories while supporting online replanning. By representing system dynamics via Hankel matrices, it integrates data-driven predictive control with a step-to-step (S2S) state transition model, synthesizing motion and responding to real-time disturbances using only historical input–output data—without requiring explicit dynamical models. Experimental validation on the Atalante exoskeleton platform demonstrates significant improvements in walking robustness and environmental adaptability, enabling stable bipedal gait under complex, time-varying conditions.

0 citationsRead paper

Leg Exoskeleton Odometry using a Limited FOV Depth Sensor

Feb 26, 2025

In lower-limb exoskeletons, narrow field-of-view (FoV) depth sensors—constrained by mechanical mounting—suffer severe motion-induced distortion, leading to significant odometry drift. To address this, we propose an EKF-ICP collaborative fusion framework that tightly integrates narrow-FoV depth point clouds with exoskeleton proprioceptive measurements (joint encoders and IMU). State estimation is performed via an extended Kalman filter (EKF), while a customized iterative closest point (ICP) algorithm enhances point cloud registration robustness under strong motion disturbances, enabling high-fidelity terrain elevation mapping. Experimental results demonstrate a 62% reduction in horizontal pose drift compared to a pure proprioceptive baseline; moreover, the generated elevation maps exhibit substantially improved completeness and geometric consistency over conventional point cloud mapping approaches.

0 citationsRead paper