Neural Networks Enhanced Optimal Admittance Control of Robot-Environment Interaction Using Reinforcement Learning

Fractional-order stochastic gradient descent method with momentum and energy for deep neural networks

In this paper, a novel fractional-order stochastic gradient descent with momentum and energy (FOSGDME) approach is proposed. Specifically, to address the challenge of converging to a real extreme point encountered by the existing fractional gradient algorithms, a novel fractional-order stochastic gradient descent (FOSGD) method is presented by modifying the definition of the Caputo fractional-order derivative. A FOSGD with moment (FOSGDM) is established by incorporating momentum information to accelerate the convergence speed and accuracy further. In addition, to improve the robustness and accuracy, a FOSGD with moment and energy is established by further introducing energy formation. The extensive experimental results on the image classification CIFAR-10 dataset obtained with ResNet and DenseNet demonstrate that the proposed FOSGD, FOSGDM and FOSGDME algorithms are superior to the integer order optimization algorithms, and achieve state-of-the-art performance.

Event-triggered adaptive neural prescribed performance admittance control for constrained robotic systems without velocity measurements

In this paper, an event-triggered adaptive neural prescribed performance admittance control (ETANPPAC) scheme is proposed to control the constrained robotic systems without velocity sensors. To ensure compliance during human-robot interaction, the reference trajectory is obtained by reshaping the desired trajectory for the robotic systems based on the admittance relationship, where a saturation function is used to constrain the reference trajectory, avoiding excessive contact forces that could render the trajectory inexecutable. Moreover, a barrier Lyapunov function is used to constrain the tracking errors for prescribed performance, where a velocity observer and a radial basis function neural network are designed to estimate the velocity and the uncertainty of the robotic systems, respectively, to enhance control performance. To reduce the communication burden, an event-triggered mechanism is introduced and the Zeno behavior is avoided with the event-triggered condition. The stability of the whole control scheme is analyzed by the Lyapunov function. Simulation and experimental tests demonstrate that the proposed ETANPPAC scheme can track the desired trajectory well under constraints and reduce the communication burden, thereby achieving better efficiency for controlling the robotic systems compared with similar control schemes.

A Novel Robotic Pushing and Grasping Method Based on Vision Transformer and Convolution

Robotic grasping techniques have been widely studied in recent years. However, it is always a challenging problem for robots to grasp in cluttered scenes. In this issue, objects are placed close to each other, and there is no space around for the robot to place the gripper, making it difficult to find a suitable grasping position. To solve this problem, this article proposes to use the combination of pushing and grasping (PG) actions to help grasp pose detection and robot grasping. We propose a pushing-grasping combined grasping network (GN), PG method based on transformer and convolution (PGTC). For the pushing action, we propose a vision transformer (ViT)-based object position prediction network pushing transformer network (PTNet), which can well capture the global and temporal features and can better predict the position of objects after pushing. To perform the grasping detection, we propose a cross dense fusion network (CDFNet), which can make full use of the RGB image and depth image, and fuse and refine them several times. Compared with previous networks, CDFNet is able to detect the optimal grasping position more accurately. Finally, we use the network for both simulation and actual UR3 robot grasping experiments and achieve SOTA performance. Video and dataset are available at https://youtu.be/Q58YE-Cc250.

Repetitive Impedance Learning-Based Physically Human-Robot Interactive Control

Model-based impedance learning control can provide variable impedance regulation for robots through online impedance learning without interaction force sensing. However, the existing related results only guarantee the closed-loop control systems to be uniformly ultimately bounded (UUB) and require the human impedance profiles being periodic, iteration-dependent, or slowly varying. In this article, a repetitive impedance learning control approach is proposed for physical human-robot interaction (PHRI) in repetitive tasks. The proposed control is composed of a proportional-differential (PD) control term, an adaptive control term, and a repetitive impedance learning term. Differential adaptation with projection modification is designed for estimating robotic parameters uncertainties in the time domain, while fully saturated repetitive learning is proposed for estimating time-varying human impedance uncertainties in the iterative domain. Uniform convergence of tracking errors is guaranteed by the PD control and the use of projection and full saturation in the uncertainties estimation and is theoretically proved based on a Lyapunov-like analysis. In impedance profiles, the stiffness and damping are composed of an iteration-independent term and an iteration- dependent disturbance, which are estimated by repetitive learning and compressed by the PD control, respectively. Therefore, the developed approach can be applied to the PHRI where iteration-dependent disturbances exist in the stiffness and damping. The control effectiveness and advantages are validated by simulations on a parallel robot in a repetitive following task.

Spatial hybrid adaptive impedance learning control for robots in repetitive interactive tasks

The existing model-based impedance learning control methods can provide variable impedance regulation for physical human-robot interaction (PHRI) in repetitive tasks without interactive force sensing, however, these methods require the completion of the repetitive tasks with constant time, which restricts their applications. For PHRI in repetitive tasks with different completion time, this paper proposes a spatial hybrid adaptive impedance learning control (SHAILC) strategy by using the spatial periodic characteristics of the tasks. In the spatial hybrid adaptation, spatial periodic adaptation is used for estimating time-varying human impedance and differential adaptation is designed for estimating robotic constant unknown parameters. The use of deadzone modifications in hybrid adaptation maintains the accuracy of the parameter estimation when the tracking error is small relative to the modeling error. The control stability is analyzed by a Lyapunov-based analysis in the spatial domain, and the control effectiveness and superiority is illustrated on a parallel robot in repetitive tasks with different task completion time.

Continuous mode adaptation for cable-driven rehabilitation robot using reinforcement learning

Continuous mode adaptation is very important and useful to satisfy the different user rehabilitation needs and improve human-robot interaction (HRI) performance for rehabilitation robots. Hence, we propose a reinforcement-learning-based optimal admittance control (RLOAC) strategy for a cable-driven rehabilitation robot (CDRR), which can realize continuous mode adaptation between passive and active working mode. To obviate the requirement of the knowledge of human and robot dynamics model, a reinforcement learning algorithm was employed to obtain the optimal admittance parameters by minimizing a cost function composed of trajectory error and human voluntary force. Secondly, the contribution weights of the cost function were modulated according to the human voluntary force, which enabled the CDRR to achieve continuous mode adaptation between passive and active working mode. Finally, simulation and experiments were conducted with 10 subjects to investigate the feasibility and effectiveness of the RLOAC strategy. The experimental results indicated that the desired performances could be obtained; further, the tracking error and energy per unit distance of the RLOAC strategy were notably lower than those of the traditional admittance control method. The RLOAC strategy is effective in improving the tracking accuracy and robot compliance. Based on its performance, we believe that the proposed RLOAC strategy has potential for use in rehabilitation robots.

Finite-Time Interactive Control of Robots with Multiple Interaction Modes

This paper proposes a finite-time multi-modal robotic control strategy for physical human-robot interaction. The proposed multi-modal controller consists of a modified super-twisting-based finite-time control term that is designed in each interaction mode and a continuity-guaranteed control term. The finite-time control term guarantees finite-time achievement of the desired impedance dynamics in active interaction mode (AIM), makes the tracking error of the reference trajectory converge to zero in finite time in passive interaction mode (PIM), and also guarantees robotic motion stop in finite time in safety-stop mode (SSM). Meanwhile, the continuity-guaranteed control term guarantees control input continuity and steady interaction modes transition. The finite-time closed-loop control stability and the control effectiveness is validated by Lyapunov-based theoretical analysis and simulations on a robot manipulator.

Robot-Agnostic Interaction Controllers Based on ROS

In robotized industrial scenarios, the need for efficiency and flexibility is increasing, especially when tasks must be executed in dangerous environments and/or require the simultaneous manipulation of dangerous/fragile objects by multiple heterogeneous robots. However, the underlying hardware and software architecture is typically characterized by constraints imposed by the robots’ manufacturers, which complicates their integration and deployment. This work aims to demonstrate that widely used algorithms for robotics, such as interaction control, can be made independent of the hardware architecture, abstraction level, and functionality provided by the low-level proprietary controllers. As a consequence, a robot-independent control framework can be devised, which reduces the time and effort needed to configure the robotic system and adapt it to changing requirements. Robot-agnostic interaction controllers are implemented on top of the Robot Operating System (ROS) and made freely available to the robotic community. Experiments were performed on the Universal Robots UR10 research robot, the Comau Smart-Six industrial robot, and their digital twins, so as to demonstrate that the proposed control algorithms can be easily deployed on different hardware and simulators without reprogramming.

Design of a Gough-Stewart Platform Based on Visual Servoing Controller

Designing a robot with the best accuracy is always an attractive research direction in the robotics community. In order to create a Gough-Stewart platform with guaranteed accuracy performance for a dedicated controller, this paper describes a novel advanced optimal design methodology: control-based design methodology. This advanced optimal design method considers the controller positioning accuracy in the design process for getting the optimal geometric parameters of the robot. In this paper, three types of visual servoing controllers are applied to control the motions of the Gough-Stewart platform: leg-direction-based visual servoing, line-based visual servoing, and image moment visual servoing. Depending on these controllers, the positioning error models considering the camera observation error together with the controller singularities are analyzed. In the next step, the optimization problems are formulated in order to get the optimal geometric parameters of the robot and the placement of the camera for the Gough-Stewart platform for each type of controller. Then, we perform co-simulations on the three optimized Gough-Stewart platforms in order to test the positioning accuracy and the robustness with respect to the manufacturing errors. It turns out that the optimal control-based design methodology helps get both the optimum design parameters of the robot and the performance of the controller {robot + dedicated controller}.

Spectral Diagnostic Model for Agricultural Robot System Based on Binary Wavelet Algorithm

The application of agricultural robots can liberate labor. The improvement of robot sensing systems is the premise of making it work. At present, more research is being conducted on weeding and harvesting systems of field robot, but less research is being conducted on crop disease and insect pest perception, nutritional element diagnosis and precision fertilizer spraying systems. In this study, the effects of the nitrogen application rate on the absorption and accumulation of nitrogen, phosphorus and potassium in sweet maize were determined. Firstly, linear, parabolic, exponential and logarithmic diagnostic models of nitrogen, phosphorus and potassium contents were constructed by spectral characteristic variables. Secondly, the partial least squares regression and neural network nonlinear diagnosis model of nitrogen, phosphorus and potassium contents were constructed by the high-frequency wavelet sensitivity coefficient of binary wavelet decomposition. The results show that the neural network nonlinear diagnosis model of nitrogen, phosphorus and potassium content based on the high-frequency wavelet sensitivity coefficient of binary wavelet decomposition is better. The R2, MRE and NRMSE of nn of nitrogen, phosphorus and potassium were 0.974, 1.65% and 0.0198; 0.969, 9.02% and 0.1041; and 0.821, 2.16% and 0.0301, respectively. The model can provide growth monitoring for sweet corn and a perception model for the nutrient element perception system of an agricultural robot, while making preliminary preparations for the realization of intelligent and accurate field fertilization.