Manipulator Arm Research Articles

A Learning-based Control Framework for Fast and Accurate Manipulation of a Flexible Object

This paper presents a learning-based control framework for fast (< 1.5 s) and accurate manipulation of a flexible object, i.e., whip targeting. The framework consists of a motion planner learned or optimized by an algorithm, Online Impedance Adaptation Control (OIAC), a sim2real mechanism, and a visual feedback component. The experimental results show that a soft actor-critic algorithm outperforms three Deep Reinforcement Learning (DRL), a nonlinear optimization, and a genetic algorithm in learning generalization of motion planning. It can greatly reduce average learning trials (to < 20%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document} of others) and maximize average rewards (to > 3 times of others). Besides, motion tracking errors are greatly reduced to 13.29%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document} and 22.36%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document} of constant impedance control by the OIAC of the proposed framework. In addition, the trajectory similarity between simulated and physical whips is 89.09%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document}. The presented framework provides a new method integrating data-driven and physics-based algorithms for controlling fast and accurate arm manipulation of a flexible object.

Quicker Path planning of a collaborative dual-arm robot using Modified BP-RRT* algorithm

Path-planning of an industrial robot is an important task to reduce the overall operation time. In industrial tasks, path planning is executed with lead-through programming, where in most cases the robot faces singulated object configurations. Cluttered environments demand path-planning algorithms, which are sensor driven, rather than pre-programmed. Path-planning algorithms, like RRT, and RRT* and their variants have inherent problems like the duration of a search and the creation of several node samples which may lead to longer path lengths. Back Propagation-Rapidly exploring Random Tree* (BP-RRT*) algorithm was a leap forward when an obstacle is enveloped with a sphere. Due to the spherical envelope of the obstacle, this method evaluates the connection between the path and obstacle in space with a spherical envelope using the triangular function and identifies the non-collision path in 3D space. This predicts the best non-collision path in the 3D workspace. The current state-of-the-art of BP-RRT* is limited to single-arm robots. A collaborative dual-arm robot faces more problems in path planning than a single-arm robot like inter-collision of manipulator arms apart from avoiding obstacles. A Modified BP-RRT* algorithm is proposed for the dual-arm collaborative robot has a pre-stage partition of grids that makes the computation faster, efficient, and collision-free compared to the traditional path planning algorithms namely RRT, RRT*, Improved RRT* and BP-RRT*. The algorithm is implemented in simulation as well as in physical implementation for ABB YuMi dual-arm collaborative robot and the typical length of the path of the proposed modified BP-RRT* method has reduced by 53.8% from the traditional RRT method, 6.95% from the RRT* method, 7.77% from improved RRT* method and 6.83% from the BP-RRT* method. Also, the average time to grasp has reduced by 17.84%, the typical duration for search has decreased by 33.45%, the number of node samples created has reduced by 14.79% from BP-RRT* algorithm.

Learning Autonomous Viewpoint Adjustment from Human Demonstrations for Telemanipulation

Teleoperation systems find many applications from earlier search-and-rescue to more recent daily tasks. It is widely acknowledged that using external sensors can decouple the view of the remote scene from the motion of the robot arm during manipulation, facilitating the control task. However, this design requires the coordination of multiple operators or may exhaust a single operator as s/he needs to control both the manipulator arm and the external sensors. To address this challenge, our work introduces a viewpoint prediction model, the first data-driven approach that autonomously adjusts the viewpoint of a dynamic camera to assist in telemanipulation tasks. This model is parameterized by a deep neural network and trained on a set of human demonstrations. We propose a contrastive learning scheme that leverages viewpoints in a camera trajectory as contrastive data for network training. We demonstrated the effectiveness of the proposed viewpoint prediction model by integrating it into a real-world robotic system for telemanipulation. User studies reveal that our model outperforms several camera control methods in terms of control experience and reduces the perceived task load compared to manual camera control. As an assistive module of a telemanipulation system, our method significantly reduces task completion time for users who choose to adopt its recommendation.

Series Clutched Actuation for Collision-Tolerant High-Speed Robots

Abstract Collisions at high-speed can severely damage robots with non-backdrivable drivetrains. Adding an overload clutch in series can improve the robot’s collision tolerance without compromising its high dynamic performance. This paper aims at determining the speed above which overload clutches are required in a two-link manipulator arm. Furthermore, the optimal clutch topology as function of the impact velocity is investigated. Third, it is evaluated if adding clutches can lower the impact force on the arm. Finally, the maximum speed is identified below which impact-aware robot control is possible. The latter requires that none of the clutches decouple during an intentional collision with the environment. These answers are obtained through collision simulations and experiments with a custom build two-link arm. It was found that adding a clutch reduces the torque experienced by the drivetrain by an order of magnitude and below the limit momentary peak torque of the strain wave gears that are used. Adding a clutch to the elbow joint of the two-link arm was effective in protecting the shoulder as well if the impact occurred at the tool center point. With respect to a rigid elbow joint, the clutched elbow joint reduced the collision force at the tool by only 8%. To demonstrate that the arm is impact-aware, a box of 8 kg is approached, impacted, and pushed at 1 m/s without decoupling a clutch, nor damaging the robot’s hardware.

Shape Control of Elastic Deformable Linear Objects for Robotic Cable Assembly

Currently, cables are manually installed in aircraft manufacturing scenarios. The utilization of robots in cable assemblies can enhance efficiency and ensure quality, presenting great promise for the future. However, the assembly process must control the shape of the cables, which is a significant challenge for robots. On the one hand, cables have high degrees of freedom in space, making accurate modeling of cable dynamics for robotic arm manipulation difficult; on the other hand, cable deformation has uncertainty, and the applied forces cause simultaneous deformation and motion, which is difficult to control. To address these problems, this study proposes a cable‐shape control method based on graph neural networks and online visual shape‐servoing. The method first approximates cable dynamics with a graph neural network. Then, in practice, the learnt model is used alongside visual‐based shape‐servoing to generate optimal robotic arm movements, controlling the cable to attain the desired shape. In the experiments, precise control of three different cable types is realized, and an example of a completed cable assembly is shown.

Safe Reinforcement Learning for Arm Manipulation with Constrained Markov Decision Process

Safe Reinforcement Learning for Arm Manipulation with Constrained Markov Decision Process

Composing MPC With LQR and Neural Network for Amortized Efficiency and Stable Control

Model predictive control (MPC) is a powerful control method that handles dynamical systems with constraints. However, solving MPC iteratively in real time, i.e., implicit MPC, remains a computational challenge. To address this, common solutions include explicit MPC and function approximation. Both methods, whenever applicable, may improve the computational efficiency of the implicit MPC by several orders of magnitude. Nevertheless, explicit MPC often requires expensive pre-computation and does not easily apply to higher-dimensional problems. Meanwhile, function approximation, although scales better with dimension, still requires pre-training on a large dataset and generally cannot guarantee to find an accurate surrogate policy, the failure of which often leads to closed-loop instability. To address these issues, we propose a triple-mode hybrid control scheme, named Memory-Augmented MPC, by combining a linear quadratic regulator, a neural network, and an MPC. From its standard form, we derive two variants of such hybrid control scheme: one customized for chaotic systems and the other for slow systems. The proposed scheme does not require pre-computation and is capable of improving the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">amortized</i> running time of the composed MPC with a well-trained neural network. In addition, the scheme maintains closed-loop stability with <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">any</i> neural networks of proper input and output dimensions, alleviating the need for certifying optimality of the neural network in safety-critical applications. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —This article was motivated by the need to reduce the amortized cost of MPC in repetitive industrial robotic applications, where long-term operational cost is important and safety is critical. Examples of such applications include factory robotic arm manipulation and fixed-route quadcopter payload transport. Unlike explicit MPC or function approximation, our approach does not require any pre-computation or pre-training. Rather, it attains task proficiency over time by learning a surrogate neural network on the spot and by gradually replacing the costly MPC with the more efficient surrogate model so long as safety permits. Consequently, the proposed scheme incurs a learning cost during the initial phase of the deployment but usually becomes more adept on the task afterwards, leading to amortized efficiency.

Unveiling the Significance of Toddler-Inspired Reward Transition in Goal-Oriented Reinforcement Learning

Toddlers evolve from free exploration with sparse feedback to exploiting prior experiences for goal-directed learning with denser rewards. Drawing inspiration from this Toddler-Inspired Reward Transition, we set out to explore the implications of varying reward transitions when incorporated into Reinforcement Learning (RL) tasks. Central to our inquiry is the transition from sparse to potential-based dense rewards, which share optimal strategies regardless of reward changes. Through various experiments, including those in egocentric navigation and robotic arm manipulation tasks, we found that proper reward transitions significantly influence sample efficiency and success rates. Of particular note is the efficacy of the toddler-inspired Sparse-to-Dense (S2D) transition. Beyond these performance metrics, using Cross-Density Visualizer technique, we observed that transitions, especially the S2D, smooth the policy loss landscape, promoting wide minima that enhance generalization in RL models.

A Concept of a Plug-In Simulator for Increasing the Effectiveness of Rescue Operators When Using Hydrostatically Driven Manipulators.

The introduction of Unmanned Ground Vehicles (UGVs) into the field of rescue operations is an ongoing process. New tools, such as UGV platforms and dedicated manipulators, provide new opportunities but also come with a steep learning curve. The best way to familiarize operators with new solutions are hands-on courses but their deployment is limited, mostly due to high costs and limited equipment numbers. An alternative way is to use simulators, which from the software side, resemble video games. With the recent expansion of the video game engine industry, currently developed software becomes easier to produce and maintain. This paper tries to answer the question of whether it is possible to develop a highly accurate simulator of a rescue and IED manipulator using a commercially available game engine solution. Firstly, the paper describes different types of simulators for robots currently available. Next, it provides an in-depth description of a plug-in simulator concept. Afterward, an example of a hydrostatic manipulator arm and its virtual representation is described alongside validation and evaluation methodologies. Additionally, the paper provides a set of metrics for an example rescue scenario. Finally, the paper describes research conducted in order to validate the representation accuracy of the developed simulator.

Machine learning meets advanced robotic manipulation

Automated industries lead to high quality production, lower manufacturing cost and better utilization of human resources. Robotic manipulator arms have major role in the automation process. However, for complex manipulation tasks, hard coding efficient and safe trajectories is challenging and time consuming. Machine learning methods have the potential to learn such controllers based on expert demonstrations. Despite promising advances, better approaches must be developed to improve safety, reliability, and efficiency of ML methods in both training and deployment phases. This survey aims to review cutting edge technologies and recent trends on ML methods applied to real-world manipulation tasks. After reviewing the related background on ML, the rest of the paper is devoted to ML applications in different domains such as industry, healthcare, agriculture, space, military, and search and rescue. The paper is closed with important research directions for future works.

Safe learning-based control for an aerial robot with manipulator arms

Aerial manipulation aims at combining the versatility and the agility of some aerial platforms as multirotor unmanned aerial vehicles (UAVs) with the manipulation capabilities of robotic arms. Trajectory tracking control of these vehicles is a substantial aspect for an increasing range of application domains. However, external disturbances and parts of the internal dynamics are often unknown or very time-consuming to model. To overcome this issue, we present a tracking control law using an learning-based oracle based on Gaussian process models. The presented approach guarantees a bounded tracking error with high probability where the bound is explicitly given. A numerical example highlights the effectiveness of the proposed control law.

Unifying the Generalized Jacobian Matrix and prioritized task hierarchies, with application to free-floating VMSs

Free-floating vehicle manipulator systems (VMSs), such as those in space or underwater, experience a coupling effect between the motion of the manipulator arm and the vehicle base, since the motion of the joints induces a motion of the base relative to the system's center of gravity (CG). In this work a control framework is proposed that takes this coupling effect into account, selecting the motion of the CG as the highest priority task in a dynamically consistent task hierarchy in order to reduce the need for counteracting disturbances while controlling the position of the manipulator workspace. The proposed approach generalizes previous works using the Generalized Jacobian matrix to allow the completion of several prioritized tracking tasks. Control allocation is performed in a manner ensuring that the thrusters are used only for controlling the overall position of the VMS, while the joints are used for tasks requiring higher accuracy. The set in which all tracking error dynamics are zero is shown to be uniformly asymptotically stable, and the performance of the proposed control method is validated in a simulation study.

Effectiveness Analysis of Hydraulic Torque Wrench Machine Using Failure Mode and Effect Analysis (FMEA) and Logic Tree Analysis Study Case on Heavy Equipment Manufacturing

Production quality in terms of process efficiency and quality in the manufacturing industry must always be improved in order to maintain customer confidence. PT XYZ as a heavy equipment assembly company is one of the companies that depends on process reliability for a smooth production process. Based on this, the problems raised in this study focus on increasing the efficiency of the process of installing bolts on slew bearings. By using Overall Equipment Efficiency (OEE) and Process Capability (CpK) as the main benchmarks for measuring the quality of the production process where the latest data shows the average OEE value is at 18.53%, while the CpK value for the bolt tightening process is at 0 ,67. The OEE and CpK figures obtained show that the process quality is still not optimal and needs to be improved. The purpose of this research is to identify and prevent as many factors as possible that can lead to process failure. The methods used to evaluate processes and to identify where and how a process might fail are Failure Mode and Effect Analysis (FMEA) and Logic Tree Analysis (LTA). Both methods are used to identify and prevent as many factors as possible that can lead to a process failure. The results of research using the FMEA and LTA methods show that in the process of installing slew bearing bolts there is a process that needs to be improved because the RPN value is quite high, namely above 125. Some suggestions for improvement such as the use of a manipulator arm on a torque tool and the implementation of a manufacturing execution system (MES) can reduce the RPN value from above 125 to 28, where the process obtained is better than before.

РОЗРОБКА МАНІПУЛЯТОРА ДЛЯ ПОВНОЇ АВТОМАТИЗАЦІЇ ПРОЦЕСУ ЗАРЯДЖАННЯ ЕЛЕКТРОМОБІЛЕЙ

Electric cars are gradually gaining popularity both in the world and in Ukraine. But in order for them to be freely used throughout the country, an appropriate infrastructure is needed. Unlike traditional cars with internal combustion engines, electric cars can be refuelled anywhere there is access to an electrical outlet. The main problems of electric cars during their operation over long distances are related to the duration of the journey on one charge and the low development of the infrastructure for them. Systems of automated charging stations are also gradually developing. There are already terminals of charging stations with payment by bank card and other payment systems, but they work on the principle of self-service. For full automation of the charging process, connecting the charging station to the electric car without human intervention is not enough. Wireless charging can solve this problem, but at this stage of development, the power transferred in this way is not always enough. To solve this problem, a combination of a manipulator arm and a typical charging station was proposed. The manipulator will determine the vehicle's position and insert the charging station's socket into the car's plug using sensors and machine vision. As part of this work, we made a conceptual project of a manipulator that will automatically connect the connector of the charging station to the car. Also, for this charging complex, a system of monitoring and protection of the electrical connection of the charger with the consumer against short circuit, overheating, sparking and arcing connector was developed, taking into account various interchangeable connectors. When developing the manipulator's arms, the following tasks were taken into account: ensuring sufficient kinematic freedom of movement to cover the working area of the charging station service and maximally simplifying the design. To reliably ensure access to any point of the working space, a kinematic scheme with two translational and three rotational links was used. Next, to determine the position of the manipulator in the workspace, the direct and inverse kinematics problems for this scheme were solved. The physical model of the manipulator along with servomotors, PID controllers and control system based on the equations of kinematics problems was simulated in the Matlab application software package. The implementation of a machine vision system and a switching device protection system is proposed for a full-fledged control system of the automated charging station. The resulting design of the automated charging station is experimental and does not have many analogues in the world. The modularity of the design ensures the versatility of its use at other types of charging stations, as well as the possibility of its modernization.

Human-Inspired Non-prehensile Manipulation Strategies: Design, Implementation, and Evaluation

Grasping of objects is not always feasible for robot manipulators, e.g., due to their geometric properties. Non-prehensile manipulation strategies can enable manipulators to successfully move these objects around. We analyze human-inspired gripper configurations for pushing small or heavy objects and propose closed-loop pushing strategies based on force-torque measurements as well as open-loop strategies to push small objects. In a thorough evaluation on a KUKA LWR4+ manipulator arm and in simulation, we discuss the effects of the different designs and strategies.

Passivity-Based Tracking Control of a Mobile Manipulator Robot

This work presents a control approach based on the passivity principle, developed to guarantee performance of the application used to track the trajectory of mobile manipulator when disturbed. By exploiting the particularity of the mobile manipulator robots modelling equipped with a nonholonomic mobile base, we present a global control law for the mobile manipulator as a single system. This control allows taking into account the whole system and modifying its dynamics by introducing a highly non-linear regressor matrix to consider uncertainties and modeling constraints. The presented controller is applied to the mobile manipulator robot composed of a manipulator's arm with 2 DDL mounted on a mobile unicycle platform. Simulation tests validate the performance of the proposed approach when external disturbances occur; showing an acceptable performance of the system's stability and validated by exploiting the Lyapunov theory.

Automatic Fruit Plucking Machine Using Deep Learning

Abstract: The Automatic Fruit Plucking Machine project aims to develop an innovative automated system to revolutionize fruit harvesting. By integrating computer vision, robotic arm manipulation, and YOLO deep learning algorithms, the machine can accurately identify and pluck ripe fruits. It reduces labor costs, enhances harvest quality, and improves productivity. The machine's core is a trained machine learning algorithm that combines image processing and deep learning models for real-time fruit detection. Through iterative testing and refinement, the machine's performance continually improves. This project offers a sustainable and efficient solution for fruit producers to meet market demands and reduce manual labor reliance

Simultaneous motion and shape control of redundant mobile manipulators

This work proposes the design of a simultaneous motion and shape control of a redundant mobile manipulator. The method is based on the construction of a variable weighting matrix that allows to control of specific sections of this type of robot composed of a mobile platform and manipulator arm. The weight matrix allows for prioritizing desired parts of the robot through cost functions that depend on an additional variable defined in this work, which can be changed automatically according to some pre-established criteria or by a human operator if the robot is teleoperated. In addition, simulations are included to show the usefulness of the proposal.

Development and Evaluation of a Felling Head for a Light Forest Crawler

With motor-manual wood harvesting (by a forest worker with a chainsaw) fatal accidents happen every year when the tree is felled or when parts of the crown fall down. The alternative is to fell trees mechanically using a timber harvester head, which, however, must be brought up to the trees in the forest by means of its crane. With the usual crane reach of 10 m, the harvester needs a system of parallel strip roads with a spacing of 20 m. Furthermore, the harvester needs a dead weight of around 20 tons that compacts the soil. Both consequences increasingly evoke critics. The requirement to fell trees mechanically and to enlarge the distance between the strip roads calls for a solution to fell trees with a small, light machine that can apply its felling tool to the tree in close proximity. Together Pfanzelt Maschinenbau GmbH and the Professorship for Forest Technology of Technische Universität Dresden have run a project for developing a compact, new type of felling head, which is attached to the existing forest crawler »Moritz FR70/75« by means of a short manipulation arm. This head imitates the felling technique, which is applied by a forest worker, in a mechanical way with a high grade of automatization. Even though this machine works with higher system costs, it is significantly faster and more precise than the motor-manual version. The functional principle of the felling head was developed, patented, conceptualized and optimized with the help of prototypes and individual tests at the TU Dresden, Professorship for Forest Technology. After that, it was completely designed, manufactured and automated in terms of control technology by the Pfanzelt company. More than 100 conifers with a felling diameter of up to 50 cm were felled safely and without any problems with the prototype. The possible integration into harvesting processes as well as the effects on the use in the forest stands were analyzed in detail. The project has shown that it is possible to fell trees in a fully mechanized way without danger for the forest worker with a machine that weights roughly a tenth of the dead weight of a conventional harvester.

FPC-BTB detection and positioning system based on optimized YOLOv5

With the aim of addressing the visual positioning problem of board-to-board (BTB) jacks during the automatic assembly of flexible printed circuit (FPC) in mobile phones, an FPC-BTB jack detection method based on the optimized You Only Look Once, version 5 (YOLOv5) deep learning algorithm was proposed in this study. An FPC-BTB jack real-time detection and positioning system was developed for the real-time target detection and pose output synchronization of the BTB jack. On that basis, a visual positioning experimental platform that integrated a UR5e manipulator arm and Hikvision industrial camera was built for BTB jack detection and positioning experiments. As indicated by the experimental results, the developed FPC-BTB jack detection and positioning system for BTB target recognition and positioning achieved a success rate of 99.677%. Its average detection accuracy reached 99.341%, the average confidence of the detected target was 91%, the detection and positioning speed reached 31.25 frames per second, and the positioning deviation was less than 0.93 mm, which conforms to the practical application requirements of the FPC assembly process.