Article
Robotics
Zhenxi Cui, Wanyu Ma, Jiewen Lai, Henry K. Chu, Yi Guo
Summary: This letter introduces an adaptive term to solve the generalization problem of coupled multiple DMPs and model deformable objects. Based on this method, the manipulation of deformable objects can be treated as a second-order system, providing greater flexibility and robustness. Simulation and experimental results demonstrate the effectiveness of this approach.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2022)
Article
Automation & Control Systems
Hyoin Kim, Changsuk Oh, Inkyu Jang, Sungyong Park, Hoseong Seo, H. Jin Kim
Summary: This paper presents a framework that reduces the number of demonstrations and generates the overall trajectory of mobile manipulators for complex missions. The proposed method segments complex demonstrations into unit motions representing sub-tasks and utilizes phase decision and Gaussian process regression to configure multiple PDMPs and improve generalization performance. Simulation results and actual experiments confirm the effectiveness of the framework.
IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING
(2022)
Article
Robotics
Ge Li, Zeqi Jin, Michael Volpp, Fabian Otto, Rudolf Lioutikov, Gerhard Neumann
Summary: Movement Primitives (MPs) are a well-known concept for representing and generating modular trajectories. There are two types of MPs: dynamics-based approaches that generate smooth trajectories, such as Dynamic Movement Primitives (DMPs), and probabilistic approaches that capture higher-order statistics, such as Probabilistic Movement Primitives (ProMPs). However, there is currently no unified MP method that can generate smooth trajectories while capturing higher-order statistics.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2023)
Article
Robotics
Matteo Saveriano, Fares J. Abu-Dakka, Aljaz Kramberger, Luka Peternel
Summary: Biological systems, including human beings, have the innate ability to perform complex tasks in a versatile and agile manner. Dynamic Movement Primitives (DMPs) provide a mathematical formulation of motor primitives and have inspired research in various areas of robotics. This paper serves as a tutorial survey, presenting DMP formulations, discussing their advantages and limitations, and providing a comprehensive review of existing literature.
INTERNATIONAL JOURNAL OF ROBOTICS RESEARCH
(2023)
Article
Automation & Control Systems
Michele Ginesi, Nicola Sansonetto, Paolo Fiorini
Summary: Dynamic Movement Primitives (DMPs) framework is widely used for learning point-to-point trajectories, but still has shortcomings in terms of function approximation methods, trajectory adjustment dependencies, and constraints on one-shot learning.
ROBOTICS AND AUTONOMOUS SYSTEMS
(2021)
Article
Computer Science, Artificial Intelligence
Michele Ginesi, Daniele Meli, Andrea Roberti, Nicola Sansonetto, Paolo Fiorini
Summary: This study addresses the challenging obstacle avoidance in Dynamic Movement Primitives by introducing velocity information into the potential function definition, leading to smoother behavior. The effectiveness of the new formulation is validated through experiments in simulated environments and with various real robots.
JOURNAL OF INTELLIGENT & ROBOTIC SYSTEMS
(2021)
Article
Robotics
Eric Huang, Xianyi Cheng, Yuemin Mao, Arnav Gupta, Matthew T. Mason
Summary: The central theme in robotic manipulation is the robot's interaction with the world through physical contact. Manipulation primitives, specific words describing physical contact and actions, are studied individually due to the challenges of nonlinear and nonsmooth physical interaction. This paper proposes a complete and general framework for autogenerating manipulation primitives, using contact modes as a means to classify and enumerate them.
INTERNATIONAL JOURNAL OF ROBOTICS RESEARCH
(2023)
Article
Chemistry, Multidisciplinary
Ang Li, Zhenze Liu, Wenrui Wang, Mingchao Zhu, Yanhui Li, Qi Huo, Ming Dai
Summary: Dynamic movement primitives (DMPs) are a robust framework for movement generation, which can be extended by adding a perturbing term for obstacle avoidance without sacrificing stability. By using the PI2 algorithm, the profiles of potentials and the parameters of the DMPs are learned simultaneously, allowing for optimization of obstacle avoidance while completing specified tasks.
APPLIED SCIENCES-BASEL
(2021)
Article
Automation & Control Systems
Francesco Iodice, Yuqiang Wu, Wansoo Kim, Fei Zhao, Elena De Momi, Arash Ajoudani
Summary: This article proposes a method for learning and robotic replication of dynamic collaborative tasks from offline videos, by decoding dynamic information from a human skeleton model and estimating and replicating the robot's Cartesian impedance profile using Gaussian Mixture Model and Gaussian Mixture Regression.
Article
Computer Science, Artificial Intelligence
Ying Zhang, Miao Li, Chenguang Yang
Summary: In this research, a broad neural network is proposed to approximate the unknown terms of the robot, aiming to address the issue of neural network failing to contain the entire input vector for optimal approximation. This method enables the reuse of learned motion controller to complete other actions without the need to relearn weight parameters.
Article
Computer Science, Artificial Intelligence
Zhenyu Lu, Ning Wang, Miao Li, Chenguang Yang
Summary: This study proposes an incremental motor skill learning, generalization, and control method based on DMP and BLS. The method can extract both ordinary skills and instant reactive skills from demonstrations. The approach includes three steps: learning ordinary skills, achieving multistylistic reactive skill learning, and integrating and generalizing the learned skills.
IEEE TRANSACTIONS ON FUZZY SYSTEMS
(2022)
Article
Automation & Control Systems
Shuo Yang, Wei Zhang, Ran Song, Jiyu Cheng, Hesheng Wang, Yibin Li
Summary: This article proposes a watch-and-act imitation learning pipeline that enables robots to learn diverse manipulations from visual demonstrations. The system consists of a captioning module for understanding the demonstration videos and a manipulation module for learning the demonstrated manipulations. Extensive experiments validate the superiority of the two modules and a real robotic arm is used for demonstrating the whole robotic imitation system.
IEEE TRANSACTIONS ON SYSTEMS MAN CYBERNETICS-SYSTEMS
(2023)
Article
Robotics
Cheng Chi, Benjamin Burchfiel, Eric Cousineau, Siyuan Feng, Shuran Song
Summary: This paper addresses the challenge of goal-conditioned dynamic manipulation of deformable objects by presenting the Iterative Residual Policy (IRP), a learning framework applicable to tasks with complex dynamics. IRP learns an implicit policy via delta dynamics, allowing it to quickly optimize actions online to reach a specified goal. The effectiveness of IRP is demonstrated on two tasks, showcasing its generalization capability.
INTERNATIONAL JOURNAL OF ROBOTICS RESEARCH
(2023)
Article
Robotics
Xinbo Yu, Peisen Liu, Wei He, Yu Liu, Qi Chen, Liang Ding
Summary: Endowing robots with human-like abilities is an important goal in robotics. This study develops a robot skill learning framework that considers both movement and impedance features, using learning from demonstration and electromyography method to achieve adaptivity in robot task execution.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2022)
Article
Automation & Control Systems
Giovanni Franzese, Leandro de Souza Rosa, Tim Verburg, Luka Peternel, Jens Kober
Summary: Performing bimanual tasks with dual robotic setups can greatly enhance their practical applications. However, synchronizing and coordinating the single-arm policies pose challenges. This article introduces the SIMPLe algorithm, which allows for teaching and correcting single or dual arm impedance policies directly from human demonstrations. It also proposes a novel graph encoding based on Gaussian process regression for policy representation, ensuring convergence to the trajectory and demonstrated goal. The algorithm's flexibility to adapt to feedback and external perturbations was demonstrated in real-world experiments.
IEEE-ASME TRANSACTIONS ON MECHATRONICS
(2023)
Article
Multidisciplinary Sciences
Nalika Ulapane, Karthick Thiyagarajan, David Hunt, Jaime Valls Miro
JOVE-JOURNAL OF VISUALIZED EXPERIMENTS
(2020)
Article
Engineering, Electrical & Electronic
Nalika Ulapane, Karthick Thiyagarajan, Jaime Valls Miro, Sarath Kodagoda
Summary: Pulsed eddy current (PEC) sensing is widely used for detecting flaws and properties of metallic test pieces, with recent applications in ferromagnetic materials thickness quantification. By using a surface representation method, a 2% improvement in accuracy can be achieved, leading to significant cost savings in structural health monitoring and infrastructure management.
IEEE SENSORS JOURNAL
(2021)
Article
Automation & Control Systems
Xingshuo Jing, Kun Qian, Xin Xu, Jishen Bai, Bo Zhou
Summary: The paper introduces a novel domain adversarial transfer network for transferring grasping skills learned from simulated environments to the real world. By utilizing generative adversarial training and task-constrained grasp candidates, shared features are extracted to effectively reduce the domain gap between different domains.
ROBOTICS AND AUTONOMOUS SYSTEMS
(2021)
Article
Chemistry, Multidisciplinary
Arturo Gil Aparicio, Jaime Valls Miro
Summary: This paper proposes a novel stochastic method to track end-effector task-space motion in an efficient manner by improving manipulability measure and motion planning, enhancing robot performance in the presence of obstacles.
APPLIED SCIENCES-BASEL
(2021)
Article
Automation & Control Systems
Kun Qian, Xin Xu, Huan Liu, Jishen Bai, Shan Luo
Summary: This paper proposes an environment-adaptive probabilistic interaction primitive method using learning-from-demonstration, which achieves proactive assistance in human-robot collaboration. The effectiveness of the proposed method is validated through experiments.
ROBOTICS AND AUTONOMOUS SYSTEMS
(2022)
Article
Robotics
Tong Yang, Jaime Valls Miro, Yue Wang, Rong Xiong
Summary: An optimal solution to the task-space tracking problem using a non-redundant manipulator is proposed. The algorithm guarantees minimum manipulator reconfigurations during task-space tracking by selecting suitable joint-space connected segments. A faster greedy strategy is suggested to increase computational efficiency while maintaining global optimality and completeness. The effectiveness of the algorithm is validated through simulation and real-world experiments.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2022)
Article
Chemistry, Analytical
Lili Bykerk, Jaime Valls Miro
Summary: This paper presents advanced methods for leak monitoring using vibro-acoustic sensors, achieving excellent leak/no-leak classification results with Convolutional Neural Networks, regardless of sensor type, manufacturer, or region.
Article
Automation & Control Systems
Yongqiang Zhao, Xingshuo Jing, Kun Qian, Daniel Fernandes Gomes, Shan Luo
Summary: In this paper, a novel tactile-motor policy learning method is proposed to generalize tubular object manipulation skills from simulation to reality. The authors introduce an attention mechanism and a task-related constraint to narrow the pixel-level domain gap for tactile tasks using an adversarial domain adaptation network. The proposed method is implemented in a Reinforcement Learning-based policy learning framework for robotic insert-and-pullout manipulation tasks, showing its generalization capability in research laboratories.
ROBOTICS AND AUTONOMOUS SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Kun Qian, Yanhui Duan, Chaomin Luo, Yongqiang Zhao, Xingshuo Jing
Summary: This article proposes a novel real-to-sim approach for unsupervised domain adaptation in object pose estimation. By preserving structural information, semantic information, and object pose, this method outperforms conventional domain adaptation methods during the transfer.
IEEE TRANSACTIONS ON COGNITIVE AND DEVELOPMENTAL SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Xin Xu, Kun Qian, Xingshuo Jing, Wei Song
Summary: This article proposes a method for robots to learn and segment manipulation subactions from human demonstration videos, achieving simultaneous segmentation and generation of instructions through a two-stream network and self-attention mechanism. By using PDDL-based skill scripts to handle underspecified or redundant demonstrations, experimental results show that the method generates more accurate instructions than state-of-the-art methods.
IEEE TRANSACTIONS ON COGNITIVE AND DEVELOPMENTAL SYSTEMS
(2023)
Article
Engineering, Electrical & Electronic
Xingshuo Jing, Kun Qian, Tudor Jianu, Shan Luo
Summary: In this article, the authors propose an adaptively correlation-attentive and task-related network for tactile image transfer. The method leverages an adaptively correlative attention mechanism and a task-related constraint loss to narrow the domain gap between simulated and real-world tactile perception tasks. Experimental results show that the proposed method outperforms the state-of-the-art in tactile image generation and classification.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Engineering, Mechanical
Lili Bykerk, Jaime Valls Miro
Summary: Leaks in water distribution networks are a major concern for utilities worldwide, causing significant water loss and costs. Active leak detection methods can help identify hidden leaks, but their success depends on the detection instruments and operator experience. Vibro-acoustic sensors are being increasingly used by water utilities for temporary structural health monitoring to improve leak detection accuracy.
Proceedings Paper
Automation & Control Systems
Tong Yang, Jaime Valls Miro, Yue Wang, Rong Xiong
Summary: This work proposes a novel algorithm to optimally partition the task space while considering various finite locations where the object may be positioned, ensuring joint-space coverage continuity with minimal liftoffs. Results from challenging the algorithm to achieve coverage of multiple objects in simulation and real tests with an industrial manipulator demonstrate the effectiveness of this planner compared to classical coverage strategies facing the same problem.
2021 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION (ICRA 2021)
(2021)
Proceedings Paper
Automation & Control Systems
Kavindie Katuwandeniya, Stefan H. Kiss, Lei Shi, Jaime Valls Miro
Summary: This work proposes a multi-modal framework to generate user intention distributions for operating a mobile vehicle. The framework learns from past observed trajectories and leverages traversability information from visual surroundings to produce future trajectories, suitable for embedding into a perception-action shared control strategy on a mobile agent. The data-driven framework significantly reduces error in predicted trajectories compared to existing strategies, demonstrating its effectiveness in practical applications.
2021 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS (IROS)
(2021)
Article
Robotics
Aime Charles Alfred Dione, Shoichi Hasegawa
Summary: This study proposes a new method to solve the kinematic hyper redundancy problem in posture control of a robotic arm with redundant degrees of freedom. By controlling strategic points along the arm, the method guides the overall motion of the arm towards the target posture. The method is capable of safely and accurately tracking target postures that are significantly different from the initial posture.
Article
Robotics
Peirang Li, Naoya Ueda, Chi Zhu
Summary: This study focuses on the traditional attendant-propelled power-assist wheelchairs (APAWs) and identifies the discomfort caused by changes in handle velocity when passing through a slope. To address this issue, a velocity compensation method is proposed and validated through simulations and experiments.
Article
Robotics
Juan Padron, Kenta Tatsuda, Kiyoshi Ohishi, Yuki Yokokura, Toshimasa Miyazaki
Summary: This paper proposes a method that takes into account real-time posture-dependent inertial variation to achieve exact dynamic compensation and independent control of each axis for industrial robots. By discretizing the state equations of the posture-variant two-inertia system model, the whole control system can be easily redeisgned at each control cycle to address the issues caused by posture changes.
