Search

This study investigates a novel approach to robotic arm control through integrating spiking neural networks with the twin delayed deep deterministic policy gradient reinforcement learning algorithm. Specifically, it presents the first application of spiking neural networks-based twin delayed deep deterministic policy gradient in 3D robotic manipulation, demonstrating its extension from traditional 2D tasks to complex 3D target-reaching scenarios with improved energy efficiency and stability. Additionally, with the inertial measurement unit data the system successfully mimics human arm movements, achieving a success rate of 0.95 among 50 trials and enabling an intuitive and accurate human–robot interaction system. This pioneering attempt highlights the feasibility of combining the biologically inspired spiking neural networks with the reinforcement learning algorithm to address the real-time challenges in high-dimensional robotic environments and advance the field of human–robot interaction systems. [ABSTRACT FROM AUTHOR]

Creativity and style in music playing originates from constraints and imperfect interactions between instruments and players. Digital and robotic systems have so far been unable to capture this naturalistic playing. Whether as an additional tool for musicians, function restoration with prosthetics, or artificial intelligence-powered systems, the physical embodiment and interactions generated are critical for expression and connection with an audience. We introduce the parametric hand, which serves as a platform to explore the generation of diverse interactions for the stylistic playing of both pianos and guitars. The hand's anatomical design and non-linear actuation are exploitable with simple kinematic modeling and synergistic actuation. This enables the modulation of two degrees of freedom for piano chord playing and guitar strumming with up to 6.6 times the variation in the signal amplitude. When only varying hand stiffness properties, we achieve capabilities similar to the variation exhibited in human strumming. Finally, we demonstrate the exploitability of behaviours with the rapid programming of posture and stiffness for sequential instrument playing, including guitar pick grasping. In summary, we highlight the utility of embodied intelligence in musical instrument playing through interactive behavioural diversity, as well as the ability to exploit behaviours over this diversity through designed behavioural robustness and synergistic actuation. [ABSTRACT FROM AUTHOR]

This paper explores the application of Inverse Reinforcement Learning (IRL) in robotics, focusing on inferring reward functions from expert demonstrations of robot arm manipulation tasks. By leveraging IRL, we aim to develop efficient and adaptable techniques for learning robust solutions to complex tasks in continuous state spaces. Our approach combines Apprenticeship Learning via IRL with Proximal Policy Optimization (PPO), expert-trajectory-based features, and the application of a reverse discount. The feature space is constructed by sampling expert trajectories to capture essential task characteristics, enhancing learning efficiency and generalizability by concentrating on critical states. To prevent the vanishing of feature expectations in goal states, we introduce a reverse discounting application to prioritize feature expectations in final states. We validate our methodology through experiments in a simple GridWorld environment, demonstrating that reverse discounting enhances the alignment of the agent's features with those of the expert. Additionally, we explore how the parameters of the proposed feature definition influence performance. Further experiments on robotic manipulation tasks using the TIAGo robot compare our approach with state-of-the-art methods, confirming its effectiveness and adaptability in complex continuous state spaces across diverse manipulation tasks. [ABSTRACT FROM AUTHOR]

Abstract Background Amyotrophic lateral sclerosis (ALS) is a lethal progressive neurodegenerative disease characterized by upper motor neuron (UMN) and lower motor neuron (LMN) involvement. Their varying degree of involvement results in a clinical heterogenous picture, making clinical assessments of UMN signs in patients with ALS often challenging. We therefore explored whether instrumented assessment using robotic manipulation could potentially be a valuable tool to study signs of UMN involvement. Methods We examined the dynamics of the wrist joint of 15 patients with ALS and 15 healthy controls using a Wristalyzer single-axis robotic manipulator and electromyography (EMG) recordings in the flexor and extensor muscles in the forearm. Multi-sinusoidal torque perturbations were applied, during which participants were asked to either relax, comply or resist. A neuromuscular model was used to study muscle viscoelasticity, e.g. stiffness (k) and viscosity (b), and reflexive properties, such as velocity, position and force feedback gains (kv, kp and kf, respectively) that dominated the responses. We further obtained clinical signs of LMN (muscle strength) and UMN (e.g. reflexes, spasticity) dysfunction, and evaluated their relation with the estimated neuromuscular model parameters. Results Only force feedback gains (kf) were elevated in patients (p = 0.033) compared to controls. Higher kf, as well as the resulting reflexive torque (Tref), were both associated with more severe UMN dysfunction in the examined arm (p = 0.040 and p

Humans typically make decisions based on past experiences and observations, while in the field of robotic manipulation, the robot's action prediction often relies solely on current observations, which tends to make robots overlook environmental changes or become ineffective when current observations are suboptimal. To address this pivotal challenge in robotics, inspired by human cognitive processes, we propose our method which integrates historical learning and multi-view attention to improve the performance of robotic manipulation. Based on a spatio-temporal attention mechanism, our method not only combines observations from current and past steps but also integrates historical actions to better perceive changes in robots' behaviours and their impacts on the environment. We also employ a mutual information-based multi-view attention module to automatically focus on valuable perspectives, thereby incorporating more effective information for decision-making. Furthermore, inspired by human visual system which processes both global context and local texture details, we have devised a method that merges semantic and texture features, aiding robots in understanding the task and enhancing their capability to handle fine-grained tasks. Extensive experiments in RLBench and real-world scenarios demonstrate that our method effectively handles various tasks and exhibits notable robustness and adaptability. [ABSTRACT FROM AUTHOR]

This paper investigates methods that leverage physical contact between a robot's structure and its environment to enhance task performance, with a primary emphasis on improving precision. Two main approaches are examined: solving the inverse kinematics problem and employing quadratic programming, which offers computational efficiency by utilizing forward kinematics. Additionally, geometrical methods are explored to simplify robot assembly and reduce the complexity of control calculations. These approaches are implemented on a physical robotic platform and evaluated in real-time applications to assess their effectiveness. Through experimental evaluation, this study aims to understand how environmental contact can be utilized to enhance performance across various conditions, offering valuable insights for practical applications in robotics. [ABSTRACT FROM AUTHOR]

Background: Amyotrophic lateral sclerosis (ALS) is a lethal progressive neurodegenerative disease characterized by upper motor neuron (UMN) and lower motor neuron (LMN) involvement. Their varying degree of involvement results in a clinical heterogenous picture, making clinical assessments of UMN signs in patients with ALS often challenging. We therefore explored whether instrumented assessment using robotic manipulation could potentially be a valuable tool to study signs of UMN involvement. Methods: We examined the dynamics of the wrist joint of 15 patients with ALS and 15 healthy controls using a Wristalyzer single-axis robotic manipulator and electromyography (EMG) recordings in the flexor and extensor muscles in the forearm. Multi-sinusoidal torque perturbations were applied, during which participants were asked to either relax, comply or resist. A neuromuscular model was used to study muscle viscoelasticity, e.g. stiffness (k) and viscosity (b), and reflexive properties, such as velocity, position and force feedback gains (kv, kp and kf, respectively) that dominated the responses. We further obtained clinical signs of LMN (muscle strength) and UMN (e.g. reflexes, spasticity) dysfunction, and evaluated their relation with the estimated neuromuscular model parameters. Results: Only force feedback gains (kf) were elevated in patients (p = 0.033) compared to controls. Higher kf, as well as the resulting reflexive torque (Tref), were both associated with more severe UMN dysfunction in the examined arm (p = 0.040 and p < 0.001). Patients with UMN symptoms in the examined arm had increased kf and Tref compared to controls (both p = 0.037). Neither of these measures was related to muscle strength, but muscle stiffness (k) was lower in weaker patients (p = 0.012). All these findings were obtained from the relaxed test. No differences were observed during the instructions comply and resist. Conclusions: This findings are proof-of-concept that instrumented assessment using robotic manipulation is a feasible technique in ALS, which may provide quantitative, operator-independent measures relating to UMN symptoms. Elevated force feedback gains, driving larger reflexive muscle torques, appear to be particularly indicative of clinically established levels of UMN dysfunction in the examined arm. [ABSTRACT FROM AUTHOR]

Germany's healthcare sector suffers from a shortage of nursing staff, and robotic solutions are being explored as a means to provide quality care. While many robotic systems have already been established in various medical fields (e.g., surgical robots, logistics robots), there are only a few very specialized robotic applications in the care sector. In this work, a multi-functional robot is applied in a hospital, capable of performing activities in the areas of transport and logistics, interactive assistance, and documentation. The service robot platform HoLLiE was further developed, with a focus on implementing innovative solutions for handling non-rigid objects, motion planning for non-holonomic motions with a wheelchair, accompanying and providing haptic support to patients, optical recognition and control of movement exercises, and automated speech recognition. Furthermore, the potential of a robot platform in a nursing context was evaluated by field tests in two hospitals. The results show that a robot can take over or support certain tasks. However, it was noted that robotic tasks should be carefully selected, as robots are not able to provide empathy and affection that are often required in nursing. The remaining challenges still exist in the implementation and interaction of multi-functional capabilities, ensuring ease of use for a complex robotic system, grasping highly heterogeneous objects, and fulfilling formal and infrastructural requirements in healthcare (e.g., safety, security, and data protection). [ABSTRACT FROM AUTHOR]

This paper presents the preliminary results of the work on a control algorithm for a two-finger gripper equipped with an electronic skin (e-skin). The e-skin measures the magnitude and location of the pressure applied to it. Contact localization allowed the development of a reliable control algorithm for robotic grasping. The main contribution of this work is the learning algorithm that adjusts the pose of the gripper during the pre-grasp approach step based on contact information. The algorithm was tested on different objects and showed comparable grasping reliability to the vision-based approach. The developed tactile sensor-rich gripper with a dedicated control algorithm may find applications in various fields, from industrial robotics to advanced interactive robots. [ABSTRACT FROM AUTHOR]

Challenges in strawberry picking made selective harvesting robotic technology very demanding. However, the selective harvesting of strawberries is a complicated robotic task forming a few scientific research questions. Most available solutions only deal with a specific picking scenario, for example, picking only a single variety of fruit in isolation. Nonetheless, most economically viable (e.g., high‐yielding and/or disease‐resistant) varieties of strawberry are grown in dense clusters. The current perception technology in such use cases is inefficient. In this work, we developed a novel system capable of harvesting strawberries with several unique features. These features allow the system to deal with very complex picking scenarios, for example, dense clusters. Our concept of a modular system makes our system reconfigurable to adapt to different picking scenarios. We designed, manufactured, and tested a patented picking head with 2.5‐degrees of freedom (two independent mechanisms and one dependent cutting system) capable of removing possible occlusions and harvesting the targeted strawberry without any contact with the fruit flesh to avoid damage and bruising. In addition, we developed a novel perception system to localize strawberries and detect their key points, picking points, and determine their ripeness. For this purpose, we introduced two new data sets. Finally, we tested the system in a commercial strawberry growing field and our research farm with three different strawberry varieties. The results show the effectiveness and reliability of the proposed system. The designed picking head was able to remove occlusions and harvest strawberries effectively. The perception system was able to detect and determine the ripeness of strawberries with 95% accuracy. In total, the system was able to harvest 87% of all detected strawberries with a success rate of 83% for all pluckable fruits. We also discuss a series of open research questions in the discussion section. [ABSTRACT FROM AUTHOR]

In this paper, we present models and planning algorithms to slide an object on a planar surface via frictional patch contact made with its top surface, whether the surface is horizontal or inclined. The core of our approach is the asymmetric dual limit surfaces model that determines slip boundary conditions for both the top and support patch contacts made with the object. This model enables us to compute a range of twists that can keep the object in sticking contact with the robot end-effector while slipping on the supporting plane. Based on these constraints, we derive a planning algorithm to slide objects with only top contact to arbitrary goal poses without slippage between end effector and the object. We fit the proposed model and demonstrate its predictive accuracy on a variety of object geometries and motions. We also evaluate the planning algorithm over a variety of objects and goals, demonstrating an orientation error improvement of 90% when compared to methods naive to linear path planners. For more results and information, please visit . [ABSTRACT FROM AUTHOR]

Creativity and style in music playing originates from constraints and imperfect interactions between instruments and players. Digital and robotic systems have so far been unable to capture this naturalistic playing. Whether as an additional tool for musicians, function restoration with prosthetics, or artificial intelligence-powered systems, the physical embodiment and interactions generated are critical for expression and connection with an audience. We introduce the parametric hand, which serves as a platform to explore the generation of diverse interactions for the stylistic playing of both pianos and guitars. The hand’s anatomical design and non-linear actuation are exploitable with simple kinematic modeling and synergistic actuation. This enables the modulation of two degrees of freedom for piano chord playing and guitar strumming with up to 6.6 times the variation in the signal amplitude. When only varying hand stiffness properties, we achieve capabilities similar to the variation exhibited in human strumming. Finally, we demonstrate the exploitability of behaviours with the rapid programming of posture and stiffness for sequential instrument playing, including guitar pick grasping. In summary, we highlight the utility of embodied intelligence in musical instrument playing through interactive behavioural diversity, as well as the ability to exploit behaviours over this diversity through designed behavioural robustness and synergistic actuation.

This research study represents a polydexterous deep reinforcement learning-based pick-and-place framework for industrial clutter scenarios. In the proposed framework, the agent tends to learn the pick-and-place of regularly and irregularly shaped objects in clutter by using the sequential combination of prehensile and non-prehensile robotic manipulations involving different robotic grippers in a completely self-supervised manner. The problem was tackled as a reinforcement learning problem; after the Markov decision process (MDP) was designed, the off-policy model-free Q-learning algorithm was deployed using deep Q-networks as a Q-function approximator. Four distinct robotic manipulations, i.e., grasp from the prehensile manipulation category and inward slide, outward slide, and suction grip from the non-prehensile manipulation category were considered as actions. The Q-function comprised four fully convolutional networks (FCN) corresponding to each action based on memory-efficient DenseNet-121 variants outputting pixel-wise maps of action-values jointly trained via the pixel-wise parametrization technique. Rewards were awarded according to the status of the action performed, and backpropagation was conducted accordingly for the FCN generating the maximum Q-value. The results showed that the agent learned the sequential combination of the polydexterous prehensile and non-prehensile manipulations, where the non-prehensile manipulations increased the possibility of prehensile manipulations. We achieved promising results in comparison to the baselines, differently designed variants, and density-based testing clutter. [ABSTRACT FROM AUTHOR]

Packaging is one of the least automated steps among all the fruit postharvest processes, which is time-consuming and labor-intensive. Therefore, a robust suction strategy for robotic manipulation needs to be developed. In this research, a geometric-based apple suction strategy for robotic packaging was studied, including suction cup design, optimal suction region selection algorithm, and robot system integration. In the first place, on the basis of the geometric features of the spheroid fruit, the structure of the suction cups was designed to provide reliable suction force. Then, suction force measurement experiments on both acrylic balls and apples were conducted. Based on the results, the parameters of the suction cup were finally determined. The results also indicated that the curvature radius of the suction region is supposed to larger than that of the suction cups. Furthermore, a robust suction region selection algorithm was developed, which involves four steps: RGB-D information acquisition, object detection and point cloud generation, spherical fitting, and suction region selection. Finally, the above methods were integrated into a robotic packaging system. In addition, on the basis of spatial-frequency domain imaging (SFDI) technology, early stage bruise was detected for validation. The results showed that, the proposed suction strategy and system is potential for robust robotic apple packaging. [ABSTRACT FROM AUTHOR]

Robotic manipulation in surgical applications often demands the surgical instrument to pivot around a fixed point, known as remote center of motion (RCM). The RCM constraint ensures that the pivot point of the surgical tool remains stationary at the incision port, preventing tissue damage and bleeding. Precisely and efficiently controlling tool positioning and orientation under this constraint poses a complex Inverse Kinematics (IK) problem that must be solved in real-time to ensure patient safety and minimize complications. To address this problem, we propose PivotIK, an efficient IK solver that combines efficient evolutionary exploration with multi-objective Jacobian-based optimization. PivotIK can track desired tool poses accurately while satisfying the RCM constraint in real-time. We evaluated PivotIK through simulations and real-world experiments using redundant robotic manipulators with multi-degree-of-freedom surgical instruments. We compare PivotIK with other IK solvers in terms of solve rates, computation times, tracking errors, and RCM errors under various scenarios, including unconstrained and RCM-constrained trajectories. Our results show that PivotIK achieves superior performance, solving the IK problem in less than 1 ms with errors below 0.01 and 0.1 mm for tracking and RCM, respectively. Our real-world experiments confirm the effectiveness of PivotIK in ensuring smooth trajectory tracking and RCM compliance. PivotIK offers a promising solution for real-time IK for robotic manipulation under RCM constraints in surgical applications. [ABSTRACT FROM AUTHOR]

Germany’s healthcare sector suffers from a shortage of nursing staff, and robotic solutions are being explored as a means to provide quality care. While many robotic systems have already been established in various medical fields (e.g., surgical robots, logistics robots), there are only a few very specialized robotic applications in the care sector. In this work, a multi-functional robot is applied in a hospital, capable of performing activities in the areas of transport and logistics, interactive assistance, and documentation. The service robot platform HoLLiE was further developed, with a focus on implementing innovative solutions for handling non-rigid objects, motion planning for non-holonomic motions with a wheelchair, accompanying and providing haptic support to patients, optical recognition and control of movement exercises, and automated speech recognition. Furthermore, the potential of a robot platform in a nursing context was evaluated by field tests in two hospitals. The results show that a robot can take over or support certain tasks. However, it was noted that robotic tasks should be carefully selected, as robots are not able to provide empathy and affection that are often required in nursing. The remaining challenges still exist in the implementation and interaction of multi-functional capabilities, ensuring ease of use for a complex robotic system, grasping highly heterogeneous objects, and fulfilling formal and infrastructural requirements in healthcare (e.g., safety, security, and data protection).

The United Nations (UN)’s call for a decade of “ecosystem restoration” was prompted by the need to address the extensive impact of anthropogenic activities on natural ecosystems. Marine ecosystem restoration is increasingly necessary due to increasing habitat degredation in deep waters (>200 m depth). At these depths, which are far beyond those accessible by divers, only established and emerging robotic platforms such as remotely operated vehicles (ROVs), autonomous underwater vehicles (AUVs), landers, and crawlers can operate through manipulators and multiparametric sensor arrays (e.g., optoacoustic imaging, omics, and environmental probes). The use of advanced technologies for deep-sea ecosystem restoration can provide: ① high-resolution three-dimensional (3D) imaging and acoustic mapping of substrates and key taxa, ② physical manipulation of substrates and key taxa, ③ real-time supervision of remote operations and long-term ecological monitoring, and ④ the potential to work autonomously. Here, we describe how robotic platforms with in situ manipulation capabilities and payloads of innovative sensors could autonomously conduct active restoration and monitoring across large spatial scales. We expect that these devices will be particularly useful in deep-sea habitats, such as ① reef-building cold-water corals, ② soft-bottom bamboo corals, and ③ soft-bottom fishery resources that have already been damaged by offshore industries (i.e., fishing and oil/gas).

This article presents a literature review of the past five years of studies using Deep Reinforcement Learning (DRL) and Inverse Reinforcement Learning (IRL) in robotic manipulation tasks. The reviewed articles are examined in various categories, including DRL and IRL for perception, assembly, manipulation with uncertain rewards, multitasking, transfer learning, multimodal, and Human-Robot Interaction (HRI). The articles are summarized in terms of the main contributions, methods, challenges, and highlights of the latest and relevant studies using DRL and IRL for robotic manipulation. Additionally, summary tables regarding the problem and solution are presented. The literature review then focuses on the concepts of trustworthy AI, interpretable AI, and explainable AI (XAI) in the context of robotic manipulation. Moreover, this review provides a resource for future research on DRL/IRL in trustworthy robotic manipulation.

Robotic manipulation in semi-structured environments require perception, planning and execution capabilities to be robust to deviations and adaptive to changes, and knowledge representation and reasoning may play a role in this direction in order to make robots aware of the situations, of the planning domains and of their own execution structures. This paper proposes an approach aimed at enhancing the perception capabilities of robotic systems through the integration of various technologies. In particular, the novelties of the proposed smart perception module include the combination of visual sensor data, object detection, and pose estimation techniques, leveraging a fiducial markers and deep learning-based methods, being able to integrate multiple sensors and perception pipelines. In addition, reasoning capabilities are introduced through the utilization of ontologies. The result is a robust and smart perception system capable of handling both simulated and real-world scenarios which in turn provides the required functionalities to allow the robot to understand its surroundings, with a primary focus on robotic manipulation tasks. The discussion on the tools used and the key implementation details are included, as well as the results in some simulated and real scenarios that validate the proposal as a module that provides situation awareness to allow a manipulation framework to adapt the robot actions to uncertain and changing scenarios.

In this manuscript, we propose a motion strategy for manipulating strings with unknown properties. Our approach iteratively refines its motion generation based on parameters estimated from observed string behavior, without the need for real-time feedback. This strategy has been shown effective in achieving several motion objectives using uniform strings of similar lengths. In this research, we improve upon this strategy by addressing the challenges posed by varying string lengths and non-uniform strings. For this, we utilize a non-uniform string model and address various string properties to demonstrate the feasibility of our proposed motion strategy. Experiments conducted with different string types and lengths (between 300 to 610mm), including some with non-uniform mass distributions, demonstrate our method’s effectiveness. Results show that our proposed method functions effectively with various kinds of strings, regardless of length and mass distribution, without requiring precise model parameters. Unique to this approach is its ability to adapt to various string characteristics through parameter estimation and motion generation, significantly reducing the need for real-world manipulation trials. Our findings illustrate the potential of our method for use in advanced robotic applications that require handling deformable objects.

Soft robotics technology has been rapidly expanding in recent years due to its advantages in flexibility and safety for human operators. Within this trend, soft grippers enable more delicate and adaptable manipulations, minimizing damage to the final object or the environment. 3D printing has recently emerged as a new manufacturing method for robotics, offering novel materials and design possibilities. The use of soft materials, such as thermoplastic elastomers (TPE) or silicone based elastomers, in 3D printing has enabled soft grippers to demonstrate their potential, leading to new applications in the medical, industrial or even the agricultural field, as well as improved performance. The ongoing synergy between soft robotics and 3D printing holds promise for continued breakthroughs, expanding the horizons of possibilities in these dynamic and evolving technological domains. This article provides a comprehensive review of the latest adavancements related to 3D printed soft grippers, as well as a discussion of the challenges ahead for this emerging field; in terms of limited resources, manufacturing costs and design process; emphasizing its growing importance in the fields of robotics and automation.

In recent years, the domain of robotic manipulation has broadened its focus from rigid objects to more complex tasks involving deformable objects. Compared to rigid bodies, the model-based manipulation of deformable objects requires online adaptation of the model, as it may change during manipulation. This challenge is also evident in the case of Defromable Linear Objects (DLOs), such as wires, hoses, or pipes. In this paper, we introduce a novel model-based method for manipulating DLO that eliminates the need for calibration between the robots and the RGBD camera. We achieve this by utilizing a local linear DLO model, represented by a Jacobian, which maps the movement of the robot’s grippers to the observed DLO displacement. We propose updating this model online using a Recursive Least Squares (RLS) adaptive filter and three distinct Jacobian update strategies. We assess the efficiency of the proposed approaches in nine different real-world manipulation scenarios using three types of wires. The experiments conducted demonstrate that all the proposed strategies provide accurate DLO shape control. However, the update strategy that employs a single action per step, combined with an intermediate DLO state prediction, emerges as the most efficient.

This paper explores the application of Inverse Reinforcement Learning (IRL) in robotics, focusing on inferring reward functions from expert demonstrations of robot arm manipulation tasks. By leveraging IRL, we aim to develop efficient and adaptable techniques for learning robust solutions to complex tasks in continuous state spaces. Our approach combines Apprenticeship Learning via IRL with Proximal Policy Optimization (PPO), expert-trajectory-based features, and the application of a reverse discount. The feature space is constructed by sampling expert trajectories to capture essential task characteristics, enhancing learning efficiency and generalizability by concentrating on critical states. To prevent the vanishing of feature expectations in goal states, we introduce a reverse discounting application to prioritize feature expectations in final states. We validate our methodology through experiments in a simple GridWorld environment, demonstrating that reverse discounting enhances the alignment of the agent’s features with those of the expert. Additionally, we explore how the parameters of the proposed feature definition influence performance. Further experiments on robotic manipulation tasks using the TIAGo robot compare our approach with state-of-the-art methods, confirming its effectiveness and adaptability in complex continuous state spaces across diverse manipulation tasks.

Agriculture 4.0 presents several challenges for the automation of various operations, including the fundamental task of harvesting. One of the crucial aspects in the automatic harvesting of high value crops is the grip and detachment of delicate fruits without spoiling them or interfering with the environment. Soft robotic systems, particularly soft grippers, offer a promising solution for this problem, as they can operate in unstructured environments, manipulate objects delicately, and interact safely with humans. In this context, this article presents a soft gripper design for harvesting as well as for pick-and- place operations of small and medium-sized fruits. The gripper is fabricated using the 3D printing technology with a flexible thermoplastic elastomer filament. This approach enables the production of an economical, compact, easily replicable, and interchangeable gripper by utilizing soft robotics principles, such as flexible structures and pneumatic actuation. [ABSTRACT FROM AUTHOR]

The control of deformable linear objects (DLOs) such as cables presents a significant challenge for robotic systems due to their unpredictable behavior during manipulation. This paper introduces a novel approach for cable shape control using dual robotic arms on a two–dimensional plane. A discrete point model is utilized for the cable, and a path generation algorithm is developed to define intermediate cable shapes, facilitating the transformation of the cable into the desired profile through a formulated optimization problem. The problem aims to minimize the discrepancy between the cable configuration and the targeted shape to ensure an accurate and stable deformation process. Moreover, a cable dynamic model is developed in which the manipulation approach is validated using this model. Additionally, the approach is tested in a simulation environment in which a framework of two manipulators grasps a cable. The results demonstrate the feasibility and accuracy of the proposed method, offering a promising direction for robotic manipulation of cables. [ABSTRACT FROM AUTHOR]

In the quest for intelligent robots, it is essential to enable them to understand tasks beyond mere manipulation. Achieving this requires a robust parsing mode that can be used to understand human cognition and semantics. However, the existing methods for task and motion planning lack generalization and interpretability, while robotic knowledge bases primarily focus on static manipulation objects, neglecting the dynamic tasks and skills. To address these limitations, we present a knowledge-based framework for hierarchically understanding various factors and knowledge types in robotic manipulation. Using this framework as a foundation, we collect a knowledge graph dataset describing manipulation tasks from text datasets and an external knowledge base with the assistance of large language models and construct the knowledge base. The reasoning tasks of entity alignment and link prediction are accomplished using a graph embedding method. A robot in real-world environments can infer new task execution plans based on experience and knowledge, thereby achieving manipulation skill transfer. [ABSTRACT FROM AUTHOR]

Today, robotic arms are widely used in industry. Reinforcement learning algorithms are used frequently for controlling robotic arms in complex environments. One of the customs off-policy model-free actor-critic deep reinforcement learning for continuous action spaces is deep deterministic policy gradient (DDPG). This algorithm has achieved significant results when applied to control robotic arms with high degrees of freedom. But, it also has limitations. DDPG is prone to instability and divergence in complex tasks due to the high dimensional continuous action spaces. In this paper, in order to increase the reliability and convergence speed of the DDPG algorithm, a new modified convergence DDPG (MCDDPG) algorithm is presented. By saving and reusing desirable parameters of the previous actor and critic networks, the proposed algorithm has shown a significant enhancement in training time and stability of the model compared to the conventional DDPG. We evaluate our method on the PR2's right arm which is a 7-DoF manipulator, and simulations demonstrate that our MCDDPG outperforms state-of-the-art algorithms such as DDPG and normalized advantage function in learning complex robotic tasks. [ABSTRACT FROM AUTHOR]

This research study represents a polydexterous deep reinforcement learning-based pick-and-place framework for industrial clutter scenarios. In the proposed framework, the agent tends to learn the pick-and-place of regularly and irregularly shaped objects in clutter by using the sequential combination of prehensile and non-prehensile robotic manipulations involving different robotic grippers in a completely self-supervised manner. The problem was tackled as a reinforcement learning problem; after the Markov decision process (MDP) was designed, the off-policy model-free Q-learning algorithm was deployed using deep Q-networks as a Q-function approximator. Four distinct robotic manipulations, i.e., grasp from the prehensile manipulation category and inward slide, outward slide, and suction grip from the non-prehensile manipulation category were considered as actions. The Q-function comprised four fully convolutional networks (FCN) corresponding to each action based on memory-efficient DenseNet-121 variants outputting pixel-wise maps of action-values jointly trained via the pixel-wise parametrization technique. Rewards were awarded according to the status of the action performed, and backpropagation was conducted accordingly for the FCN generating the maximum Q-value. The results showed that the agent learned the sequential combination of the polydexterous prehensile and non-prehensile manipulations, where the non-prehensile manipulations increased the possibility of prehensile manipulations. We achieved promising results in comparison to the baselines, differently designed variants, and density-based testing clutter.

In this paper, we propose Multi-View Dreaming, a novel reinforcement learning agent for integrated recognition and control from multi-view observations by extending Dreaming. Most current reinforcement learning method assumes a single-view observation space, and this imposes limitations on the observed data, such as lack of spatial information and occlusions. This makes obtaining ideal observational information from the environment difficult and is a bottleneck for real-world robotics applications. In this paper, we use contrastive learning to train a shared latent space between different viewpoints and show how the Products of Experts approach can be used to integrate and control the probability distributions of latent states for multiple viewpoints. We also propose Multi-View DreamingV2, a variant of Multi-View Dreaming that uses a categorical distribution to model the latent state instead of the Gaussian distribution. Experiments show that the proposed method outperforms simple extensions of existing methods in a realistic robot control task. [ABSTRACT FROM AUTHOR]

Generating motions to grasp the objects in the scene is critical in robotic manipulation. The feasibility and quality of the output trajectory are closely related to both the grasping pose and the motion to reach the object, and decoupling grasp and motion planning cannot balance these two factors. In this paper, a motion planning method is proposed that comprehensively considers trajectory optimization, trajectory goal selection, and grasp adjustment. Trajectory optimization is formulated as an equality-constrained optimization problem, in which the end configuration of the trajectory can be changed during optimization. The update rule is derived and acts as a basic rule for goal selection and adjustment. The grasp configuration is chosen from a goal set, which helps the optimization to jump out of the local minima and generate smooth and feasible trajectories. We propose a two-stage grasp selection algorithm to balance between exploitation of the trajectory optimization and exploration of the goal set. To further optimize the trajectory, we propose a grasp adjustment method that allows the grasp pose to rotate about the connection line of the contact points in the workspace. To address this complicated problem, we divide the problem into two subproblems: computing the angle to rotate in the workspace and calculating the grasp configuration in the joint space. We design the overall planner and incorporate our method into a complete robotic arm platform with perception and the grasp generation network. All proposed methods are validated by performing comparative simulations and experiments. [ABSTRACT FROM AUTHOR]