The Application of Connectionist Structures to Learning Impedance Control in Robotic Contact Tasks

The goal of this paper is to consider the synthesis of learning impedance control using recurrent connectionist structures for on-line learning of robot dynamic uncertainties in the case of robot contact tasks. The connectionist structures are integrated in non-learning impedance control laws that are intended to improve the transient dynamic response immediately after the contact. The recurrent neural network as a part of hybrid learning control algorithms uses fast learning rules and available sensor information in order to improve the robotic performance progressively for a minimum possible number of learning epochs. Some simulation results of deburring process with the MANUTEC r3 robot are presented here in order to verify the effectiveness of the proposed control learning algorithms.     

A neural network-based classification of environment dynamicsmodels for compliant control of manipulation robots

In this paper, a new method for selecting the appropriate compliance control parameters for robot machining tasks based on connectionist classification of unknown dynamic environments, is proposed. The method classifies the type of environment by using multilayer perceptron, and then, determines the control parameters for compliance control using the estimated characteristics. An important feature is that the process of pattern association can work in an on-line mode as a part of selected compliance control algorithm. Convergence process is improved by using evolutionary approach (genetic algorithms) in order to choose the optimal topology of the proposed multilayer perceptron. Compliant motion simulation experiments with robotic arm placed in contact with dynamic environment, described by the stiffness model and by the general impedance model, have been performed in order to verify the proposed approach.     

Reinforcement learning based on movement primitives for contact tasks

Recently, robot learning through deep reinforcement learning has incorporated various robot tasks through deep neural networks, without using specific control or recognition algorithms. However, this learning method is difficult to apply to the contact tasks of a robot, due to the exertion of excessive force from the random search process of reinforcement learning. Therefore, when applying reinforcement learning to contact tasks, solving the contact problem using an existing force controller is necessary. A neural-network-based movement primitive (NNMP) that generates a continuous trajectory which can be transmitted to the force controller and learned through a deep deterministic policy gradient (DDPG) algorithm is proposed for this study. In addition, an imitation learning algorithm suitable for NNMP is proposed such that the trajectories similar to the demonstration trajectory are stably generated. The performance of the proposed algorithms was verified using a square peg-in-hole assembly task with a tolerance of 0.1 mm. The results confirm that the complicated assembly trajectory can be learned stably through NNMP by the proposed imitation learning algorithm, and that the assembly trajectory is improved by learning the proposed NNMP through the DDPG algorithm.     

Robust Motion Control for Robotic Assembly Tasks using Variable Structure Control Scheme

The success of robot assembly tasks depends heavily on its ability to handle the interactions which take place between the parts being assembled. In this paper, a robust motion-control method is presented for robot manipulators performing assembly tasks in the presence of dynamic constraints from the environment. Using variable structure model reaching control concept, the control objectives is first formulated as a performance model in the task space. A dynamic compensator is then introduced to form the switching function such that the sliding-mode matches the desired model. A simple variable structure control law is suggested to force the system to reach and stay on the sliding mode so that the specified model is achieved.The proposed method is applied to control the prismatic joint of a selective compliance assembly robot-arm type robot for the insertion of printed circuit board into an edge connector socket. Various amounts of interaction forces are generated during the operation. Experimental and simulation results demonstrated the performance of the variable structure model reaching control approach. In comparison, it is shown that the popular position controllers such as proportional plus derivative control and proportional plus derivative with model-based feedforward control are not suitable for achieving good trajectory tracking accuracy in assembly tasks which experience potential interaction force.     

Adaptive robust control and admittance control for contact-driven robotic surface conditioning

This work presents a hybrid position/force control of robots for surface contact conditioning tasks such as polishing, profiling, deburring, etc. The robot force control is designed using sliding mode ideas to benefit from robustness. On the one hand, a set of equality constraints are defined to attain the desired tool pressure on the surface, as well as to keep the tool orientation perpendicular to the surface. On the other hand, inequality constraints are defined to adapt the tool position to unmodeled features present in the surface, e.g., a protruding window frame. Conventional and non-conventional sliding mode controls are used to fulfill the equality and inequality constraints, respectively. Furthermore, in order to deal with sudden changes of the material stiffness, which are forwarded to the robot tool and can produce instability and bad performance, adaptive switching gain laws are considered not only for the conventional sliding mode control but also for the non-conventional sliding mode control. A lower priority tracking controller is also defined to follow the desired reference trajectory on the target surface. Moreover, the classical admittance control typically used in force control tasks is adapted for the proposed surface contact application in order to experimentally compare the performance of both control approaches. The effectiveness of the proposed method is substantiated by experimental results using a redundant 7R manipulator, whereas its advantages over the classical admittance control approach are experimentally shown.     

基于神经网络的机械手示教系统研制

为了提升自动化机械智能水平，推动高新技术向生产力的转化，该文介绍了基于四层感知器神经网络的机械手示教系统的结构和应用实例，说明了四层感知器神经网络的设计、有导师学习的工作原理，以及针对含有不可微函数环节的神经网络所采用的综合反向传 播法。     

Syntactic systematicity in sentence processing with a recurrent self-organizing network

As potential candidates for explaining human cognition, connectionist models of sentence processing must demonstrate their ability to behave systematically, generalizing from a small training set. It has recently been shown that simple recurrent networks and, to a greater extent, echo-state networks possess some ability to generalize in artificial language learning tasks. We investigate this capacity for a recently introduced model that consists of separately trained modules: a recursive self-organizing module for learning temporal context representations and a feedforward two-layer perceptron module for next-word prediction. We show that the performance of this architecture is comparable with echo-state networks. Taken together, these results weaken the criticism of connectionist approaches, showing that various general recursive connectionist architectures share the potential of behaving systematically.     

Convergence and robustness of a discrete-time learning control scheme for constrained manipulators

The constrained motion control is one of the most common control tasks found in many industrial robot applications. The nonlinear and nonclassical nature of the dynamic model of constrained robots make designing a controller for accurate tracking of both motion and force a difficult problem. In this article, a discrete-time learning control problem for precise path tracking of motion and force for constrained robots is formulated and solved. The control system is able to reduce the tracking error iteratively in the presence of external disturbances and errors in initial condition as the robot repeats its action. Computer simulation result is presented to demonstrate the performance of the proposed learning controller. © 1994 John Wiley & Sons, Inc.     

Dynamic control of a robot manipulator for high-performance conveyor tracking

A new robot control scheme for the specific application to conveyor tracking is presented. To improve the performance of conveyor tracking, the robot arm dynamics is incorporated into the control scheme. The tracking problem for the workpiece on a variable-speed conveyor is formulated as a stochastic optimal tracking problem with specified criteria. Dividing the conveyor speed into the nominal term and the perturbed term, a two-stage control strategy is employed to cope with the nonlinearity and uncertainty of the robot-conveyor system. Simulation results are given to verify the good tracking performance with fast cycle time and high accuracy obtained in a robotic workcell. © 1995 John Wiley & Sons, Inc.     

受限机器人神经形态的变结构位置/力控制

本文基于Ｊｅａｎ和Ｆｕ（１９９３）建立的受限机器人模型的降型阶形式，利用变结构系统理论，设计了具有未知动态的受限机器人轨道／力追踪控制，提出的学习方法仅仅利用了机器人动态模型的一般结构，不需要其精确信息，计算迅速，易于实现，仿真结果验证了提出的方法的有效性。     

Adaptive neural controller for cooperative multiple robot manipulator system manipulating a single rigid object

In this article, an adaptive neural controller is developed for cooperative multiple robot manipulator system carrying and manipulating a common rigid object. In coordinated manipulation of a single object using multiple robot manipulators simultaneous control of the object motion and the internal force exerted by manipulators on the object is required. Firstly, an integrated dynamic model of the manipulators and the object is derived in terms of object position and orientation as the states of the derived model. Based on this model, a controller is proposed that achieves required trajectory tracking of the object as well as tracking of the desired internal forces arising in the system. A feedforward neural network is employed to learn the unknown dynamics of robot manipulators and the object. It is shown that the neural network can cope with the unknown nonlinearities through the adaptive learning process and requires no preliminary offline learning. The adaptive learning algorithm is derived from Lyapunov stability analysis so that both error convergence and tracking stability are guaranteed in the closed loop system. Finally, simulation studies and analysis are carried out for two three-link planar manipulators moving a circular disc on specified trajectory.     

Robot Control Optimization Using Reinforcement Learning

Conventional robot control schemes are basically model-based methods. However, exact modeling of robot dynamics poses considerable problems and faces various uncertainties in task execution. This paper proposes a reinforcement learning control approach for overcoming such drawbacks. An artificial neural network (ANN) serves as the learning structure, and an applied stochastic real-valued (SRV) unit as the learning method. Initially, force tracking control of a two-link robot arm is simulated to verify the control design. The simulation results confirm that even without information related to the robot dynamic model and environment states, operation rules for simultaneous controlling force and velocity are achievable by repetitive exploration. Hitherto, however, an acceptable performance has demanded many learning iterations and the learning speed proved too slow for practical applications. The approach herein, therefore, improves the tracking performance by combining a conventional controller with a reinforcement learning strategy. Experimental results demonstrate improved trajectory tracking performance of a two-link direct-drive robot manipulator using the proposed method.     

Intelligent robotic gripper with adaptive grasping force

The on-off control robot gripper is widely employed in pick-and-place operations in Cartesian space for handling hard objects between two positions. Without contact force monitoring, it can not be applied in fragile or soft objects handling. Although, an appropriate grasping force or gripper opening for each target could be searched by trial-and-error process, it needs expensive force/torque sensor or an accurate gripper position controller. It has too expensive and complex control strategy disadvantages for most of industrial applications. In addition, it can not overcome the target slip problem due to mass uncertainty and dynamic factor. Here, an intelligent gripper is designed with embedded distributed control structure for overcoming the uncertainty of object’s mass and soft/hard features. A communication signal is specified to integrate both robot arm and gripper control kernels for executing the robotic position control and gripper force control functions in sequence. An efficient model-free intelligent fuzzy sliding mode control strategy is employed to design the position and force controllers of gripper, respectively. Experimental results of pick-and-place soft and hard objects with grasping force auto-tuning and anti-slip control strategy are shown by pictures to verify the dynamic performance of this distributed control system. The position and force tracking errors are less than 1 mm and 0.1 N, respectively.

     

Multi‐objective learning control for robotic manipulator

Several types of learning controllers have been proposed in the literature to improve the tracking performance of robot manipulators. In most cases, the learning algorithms emphasize mainly on a single objective of learning a desired motion of the end‐effector. In some applications, more than one objective may be specified at the same time. For example, a robot may be required to follow a desired trajectory (primary objective) and at the same time avoid an obstacle (secondary objective). Thus, multi‐objective learning control can be more effective to realize the collision‐free tasks. In this paper, a multi‐objective learning control problem is formulated and solved. In the proposed learning control system, the primary objective is to track a desired end‐effector's motion and several secondary objectives can be specified for the desired orientation and for obstacles avoidance. To avoid obstacles in the workspace, a new learning concept called “region learning control” is also proposed in this paper. The proposed learning controllers do not require the exact knowledge of robot kinematics and dynamics. Sufficient condition is presented to guarantee the convergence of the learning system. The proposed learning controllers are applied to a four‐link planar redundant manipulator and simulation results are presented to illustrate the performance. © 2004 Wiley Periodicals, Inc.     

Precise and smooth contact force control for a hybrid mobile robot used in polishing

Contact force is dominant in robotic polishing since it directly determines the material removal. However, due to the position and stiffness disturbance of mobile robotic polishing and the nonlinear contact process between the robot and workpiece, how to realize precise and smooth contact force control of the hybrid mobile polishing robot remains challenging. To solve this problem, the force tracking error is investigated, which indicates that the force overshoot mainly comes from the input step signal and the environmental disturbance causes force tracking error in stable state. Accordingly, an integrated contact force control method is proposed, which combines feedforward of the desired force and adaptive variable impedance control. The nonlinear tracking differentiator is used to smooth the input step signal of the desired force for force overshoot reduction. Through modeling of the force tracking error, the adaptive law of the damping parameter is established to compensate disturbance. After theoretical analysis and simulation verification, the polishing experiment is carried out. The improvement in force control accuracy and roughness of the polished surface proves the effectiveness of the proposed method. Sequentially, the proposed method is employed in the polishing of a 76-meter wind turbine blade. The measurement result indicates that the surface roughness after mobile robotic polishing is better than Ra1.6. The study provides a feasible approach to improve the polishing performance of the hybrid mobile polishing robot.     

Experimental Implementation of Impedance Based Control Schemes for Assembly Task

Compliant manipulation tasks require the robot to follow a motion trajectory and to exert a force profile while making compliant contact with a dynamic environment. For this purpose, a generalized impedance in the task space is introduced such that the desired motion and the desired interaction force can be commanded and controlled simultaneously. Several control schemes which place different emphases on motion control or force control can be derived from the generalized impedance. The impedance-based control schemes are implemented and the performance evaluated on a common test-bed which involves the insertion of a printed circuit board into an edge connector socket. Experimental results demonstrate the superior motion and force tracking ability of the generalized impedance control method. Furthermore, safe task execution can be achieved in the presence of abnormal operating situation.     

Object handling control among two-wheel robots and a human operator: An empirical approach

This article presents object handling control between two-wheel robot manipulators, and a two-wheel robot and a human operator. The two-wheel robot has been built for serving humans in the indoor environment. It has two wheels to maintain balance and is able to make contact with a human operator via an object. A position-based impedance force control method is applied to maintain stable object-handling tasks. As the human operator pushes and pulls the object, the robot also reacts to maintain contact with the object by pulling and pushing against the object to regulate a specified force. Master and slave configuration of two-wheel robots is formed for handling an object, where the master robot or a human leads the slave robot equipped with a force sensor. Switching control from position to force or vice versa is presented. Experimental studies are performed to evaluate the feasibility of the object-handling task between two-wheel mobile robots, and the robot and a human operator.     

Incremental online sparsification for model learning in real-time robot control

For many applications such as compliant, accurate robot tracking control, dynamics models learned from data can help to achieve both compliant control performance as well as high tracking quality. Online learning of these dynamics models allows the robot controller to adapt itself to changes in the dynamics (e.g., due to time-variant nonlinearities or unforeseen loads). However, online learning in real-time applications - as required in control - cannot be realized by straightforward usage of off-the-shelf machine learning methods such as Gaussian process regression or support vector regression. In this paper, we propose a framework for online, incremental sparsification with a fixed budget designed for fast real-time model learning. The proposed approach employs a sparsification method based on an independence measure. In combination with an incremental learning approach such as incremental Gaussian process regression, we obtain a model approximation method which is applicable in real-time online learning. It exhibits competitive learning accuracy when compared with standard regression techniques. Implementation on a real Barrett WAM robot demonstrates the applicability of the approach in real-time online model learning for real world systems.     

A nodal link perceptron network with applications to control of anonholonomic system

A new perceptron neural network (PNN) for functional approximation and control of a general class of nonlinear systems is introduced. The basic structure of the network along with the conditions for its exponential convergence under a suitable training law are derived. A novel discrete-time control strategy is formulated that employs the PNN for direct online estimation of the feedforward control input. The developed controller can be applied to both discrete- and continuous-time plants. Unlike most of the existing direct adaptive or learning schemes, the nonlinear plant is not assumed to be feedback linearizable. The developed controller is then applied for tracking control of a nonholonomic (free-flying) robot. The simulation results of this application demonstrate a perfect tracking performance after the network is fully trained.     

Reinforcement Learning and Robust Control for Robot Compliance Tasks

The complexity in planning and control of robot compliance tasks mainly results from simultaneous control of both position and force and inevitable contact with environments. It is quite difficult to achieve accurate modeling of the interaction between the robot and the environment during contact. In addition, the interaction with the environment varies even for compliance tasks of the same kind. To deal with these phenomena, in this paper, we propose a reinforcement learning and robust control scheme for robot compliance tasks. A reinforcement learning mechanism is used to tackle variations among compliance tasks of the same kind. A robust compliance controller that guarantees system stability in the presence of modeling uncertainties and external disturbances is used to execute control commands sent from the reinforcement learning mechanism. Simulations based on deburring compliance tasks demonstrate the effectiveness of the proposed scheme.