Selected work

Representative Papers

Shared Object Manipulation with a Team of Collaborative Quadrupeds

Oct 01, 2025

Multi-quadruped robotic manipulation of rigid objects faces challenges including limited workspace, complex force–motion coupling, and poor adaptability to dynamic environments. Method: This paper proposes a distributed cooperative control framework tailored for legged manipulators. Extending classical hybrid position–force control, the framework integrates force-closure grasping strategies with a consensus-based distributed coordination algorithm to enable real-time motion–force co-allocation under dynamic conditions. Unlike conventional fixed-base multi-arm systems, it fully exploits quadrupeds’ omnidirectional mobility and dexterous leg-based manipulation capabilities to overcome workspace constraints. Contribution/Results: Simulation and physical experiments with three Unitree A1 quadrupeds demonstrate stable collaborative grasping, transportation, and obstacle avoidance of a shared object. The system exhibits strong robustness and environmental adaptability, establishing a novel paradigm for swarm-level cooperative manipulation in unstructured or field environments.

0 citationsRead paper

Riemannian Time Warping: Multiple Sequence Alignment in Curved Spaces

Jun 02, 2025

Existing time-warping methods are primarily designed for Euclidean spaces and struggle to align multi-time-series signals embedded in Riemannian manifolds—such as robotic trajectories and orientations. This work introduces RTW, the first general-purpose Riemannian time-warping framework for multiple sequences. RTW explicitly incorporates manifold geometry by leveraging geodesic distances, dynamic programming, and manifold-constrained optimization, supporting canonical manifolds including the unit quaternion sphere and the space of symmetric positive-definite (SPD) matrices. Theoretically grounded and computationally efficient, RTW achieves significantly higher average alignment accuracy and downstream classification accuracy than state-of-the-art methods on both synthetic benchmarks and real-world KUKA LBR iiwa robot motion data. By operating intrinsically on Riemannian manifolds, RTW overcomes the fundamental limitation of conventional time-warping approaches, which are restricted to Euclidean domains.

0 citationsRead paper

SLAM&Render: A Benchmark for the Intersection Between Neural Rendering, Gaussian Splatting and SLAM

Apr 18, 2025

Existing benchmarks at the intersection of SLAM and neural rendering lack comprehensive support for temporal modeling, multimodal perception, and cross-view/lighting generalization. Method: We propose SLAM&Render—the first standardized benchmark for this interdisciplinary domain—comprising 40 synchronized multimodal sequences (RGB, depth, IMU, robot kinematics, and ground-truth poses) across five scene categories, four lighting conditions, and object rearrangements. It uniquely integrates SLAM’s temporal robustness and multi-sensor constraints with neural rendering’s viewpoint and illumination generalization requirements, and introduces robot kinematics data for the first time to enable robotic-arm SLAM evaluation. High-precision motion capture, structured scenes, and perturbed trajectories ensure strict train/test separation. Results: Experiments expose critical performance bottlenecks of NeRF, Gaussian Splatting, and related methods on coupled SLAM-rendering tasks, establishing a reproducible, extensible framework for joint evaluation.

0 citationsRead paper

Making Sense of Touch: Unsupervised Shapelet Learning in Bag-of-words Sense

Feb 06, 2025

Addressing the challenge of unsupervised long-time-series shapelet learning, this paper proposes NN-STNE—a neural network for end-to-end modeling of shapelet membership probabilities and low-dimensional representation of tactile time-series data. Methodologically, it integrates t-SNE into a hidden layer to mitigate crowding in low-dimensional embeddings; employs L1 regularization for adaptive shapelet length optimization; introduces a Gaussian-kernel-weighted MSE loss to preserve local structure; and leverages K-means initialization to accelerate convergence. The framework constructs a Bag-of-Words tactile representation grounded in shapelet membership. Evaluated on the UCR time-series benchmark and robotic tasks involving electrical component manipulation (e.g., switch actuation), NN-STNE achieves significantly higher clustering accuracy than state-of-the-art unsupervised feature learning methods.

0 citationsRead paper

Robot Cell Modeling via Exploratory Robot Motions

Feb 03, 2025

In dynamic industrial environments, conventional obstacle modeling relies heavily on external sensors, leading to high costs and poor robustness. Method: This paper proposes a purely data-driven environmental modeling approach that utilizes only robot joint encoder data. By analyzing exploratory motion trajectories, it computes swept volumes in joint space and constructs a conservative, reliable implicit geometric mesh model—eliminating the need for CAD models, external sensors, or manual calibration. The method integrates kinematic analysis, implicit surface reconstruction, and lightweight mesh generation, and is natively compatible with ROS/MoveIt. Results: Evaluated on a KUKA LBR iisy platform, the method completes environment exploration in 3 minutes and generates a usable model within 4 minutes. Collision detection error remains below 5 mm, and deployment time is reduced by over 90%. This significantly enhances rapid production-line reconfiguration capability and broadens deployment applicability across diverse industrial settings.

0 citationsRead paper

Recent publications

Latest Papers

Shared Object Manipulation with a Team of Collaborative Quadrupeds

Oct 01, 2025

Multi-quadruped robotic manipulation of rigid objects faces challenges including limited workspace, complex force–motion coupling, and poor adaptability to dynamic environments. Method: This paper proposes a distributed cooperative control framework tailored for legged manipulators. Extending classical hybrid position–force control, the framework integrates force-closure grasping strategies with a consensus-based distributed coordination algorithm to enable real-time motion–force co-allocation under dynamic conditions. Unlike conventional fixed-base multi-arm systems, it fully exploits quadrupeds’ omnidirectional mobility and dexterous leg-based manipulation capabilities to overcome workspace constraints. Contribution/Results: Simulation and physical experiments with three Unitree A1 quadrupeds demonstrate stable collaborative grasping, transportation, and obstacle avoidance of a shared object. The system exhibits strong robustness and environmental adaptability, establishing a novel paradigm for swarm-level cooperative manipulation in unstructured or field environments.

0 citationsRead paper

Riemannian Time Warping: Multiple Sequence Alignment in Curved Spaces

Jun 02, 2025

Existing time-warping methods are primarily designed for Euclidean spaces and struggle to align multi-time-series signals embedded in Riemannian manifolds—such as robotic trajectories and orientations. This work introduces RTW, the first general-purpose Riemannian time-warping framework for multiple sequences. RTW explicitly incorporates manifold geometry by leveraging geodesic distances, dynamic programming, and manifold-constrained optimization, supporting canonical manifolds including the unit quaternion sphere and the space of symmetric positive-definite (SPD) matrices. Theoretically grounded and computationally efficient, RTW achieves significantly higher average alignment accuracy and downstream classification accuracy than state-of-the-art methods on both synthetic benchmarks and real-world KUKA LBR iiwa robot motion data. By operating intrinsically on Riemannian manifolds, RTW overcomes the fundamental limitation of conventional time-warping approaches, which are restricted to Euclidean domains.

0 citationsRead paper

SLAM&Render: A Benchmark for the Intersection Between Neural Rendering, Gaussian Splatting and SLAM

Apr 18, 2025

Existing benchmarks at the intersection of SLAM and neural rendering lack comprehensive support for temporal modeling, multimodal perception, and cross-view/lighting generalization. Method: We propose SLAM&Render—the first standardized benchmark for this interdisciplinary domain—comprising 40 synchronized multimodal sequences (RGB, depth, IMU, robot kinematics, and ground-truth poses) across five scene categories, four lighting conditions, and object rearrangements. It uniquely integrates SLAM’s temporal robustness and multi-sensor constraints with neural rendering’s viewpoint and illumination generalization requirements, and introduces robot kinematics data for the first time to enable robotic-arm SLAM evaluation. High-precision motion capture, structured scenes, and perturbed trajectories ensure strict train/test separation. Results: Experiments expose critical performance bottlenecks of NeRF, Gaussian Splatting, and related methods on coupled SLAM-rendering tasks, establishing a reproducible, extensible framework for joint evaluation.

0 citationsRead paper

Making Sense of Touch: Unsupervised Shapelet Learning in Bag-of-words Sense

Feb 06, 2025

Addressing the challenge of unsupervised long-time-series shapelet learning, this paper proposes NN-STNE—a neural network for end-to-end modeling of shapelet membership probabilities and low-dimensional representation of tactile time-series data. Methodologically, it integrates t-SNE into a hidden layer to mitigate crowding in low-dimensional embeddings; employs L1 regularization for adaptive shapelet length optimization; introduces a Gaussian-kernel-weighted MSE loss to preserve local structure; and leverages K-means initialization to accelerate convergence. The framework constructs a Bag-of-Words tactile representation grounded in shapelet membership. Evaluated on the UCR time-series benchmark and robotic tasks involving electrical component manipulation (e.g., switch actuation), NN-STNE achieves significantly higher clustering accuracy than state-of-the-art unsupervised feature learning methods.

0 citationsRead paper

Robot Cell Modeling via Exploratory Robot Motions

Feb 03, 2025

In dynamic industrial environments, conventional obstacle modeling relies heavily on external sensors, leading to high costs and poor robustness. Method: This paper proposes a purely data-driven environmental modeling approach that utilizes only robot joint encoder data. By analyzing exploratory motion trajectories, it computes swept volumes in joint space and constructs a conservative, reliable implicit geometric mesh model—eliminating the need for CAD models, external sensors, or manual calibration. The method integrates kinematic analysis, implicit surface reconstruction, and lightweight mesh generation, and is natively compatible with ROS/MoveIt. Results: Evaluated on a KUKA LBR iisy platform, the method completes environment exploration in 3 minutes and generates a usable model within 4 minutes. Collision detection error remains below 5 mm, and deployment time is reduced by over 90%. This significantly enhances rapid production-line reconfiguration capability and broadens deployment applicability across diverse industrial settings.

0 citationsRead paper