Incremental few-shot learning is highly expected for practical robotics applications. On one hand, robot is desired to learn new tasks quickly and flexibly using only few annotated training samples; on the other hand, such new additional tasks should be learned in a continuous and incremental manner without forgetting the previous learned knowledge dramatically. In this work, we propose a novel Class-Incremental Few- Shot Object Detection (CI-FSOD) framework that enables deep object detection network to perform effective continual learning from just few-shot samples without re-accessing the previous training data. We achieve this by equipping the widely-used Faster-RCNN detector with three elegant components. Firstly, to best preserve performance on the pre-trained base classes, we propose a novel Dual-Embedding-Space (DES) architecture which decouples the representation learning of base and novel categories into different spaces. Secondly, to mitigate the catastrophic forgetting on the accumulated novel classes, we propose a Sequential Model Fusion (SMF) method, which is able to achieve long-term memory without additional storage cost. Thirdly, to promote inter-task class separation in feature space, we propose a novel regularization technique that extends the classification boundary further away from the previous classes to avoid misclassification. Overall, our framework is simple yet effective and outperforms the previous SOTA with a significant margin of 2.4 points in AP performance.

Universal grasping for a diverse range of objects is a challenging problem in robotics, especially in the presence of mixed properties with fragile/rigid and heavy/light. Toward universal grasping, this article presents a practical and systematic grasping control framework that enables a variable stiffness gripper to handle the objects with diverse properties using a category-aware force regulation approach, termed classification-based force grasping. Under this framework, a convolutional neural network (CNN) is employed to classify the category of the grasping object, and a grasping force is determined based on the classified category through a database that records a predefined force magnitude per category. Sequentially, the gripper can be adjusted to a force-optimized stiffness, which facilitates the achievement of an accurate grasping force regulation in a large range. Technically, two novel enabling modules are developed for grasping classification and execution, respectively. First, a novel weight imprinting technique based on center-guided feature embedding is proposed for object classification. It enables the CNN to efficiently handle novel object categories using only a few samples even without retraining/fine-tuning. Second, a vision-based grasping force sensing module is developed, which takes advantage of the specifically designed variable-stiffness gripper. Its grasping force can be estimated from the deflection angle of finger flexure by the vision so that the contact force can be sensed and regulated. Remarkably, only single-source vision information is needed for both of the above modules without any additional force sensor. Experiments are conducted extensively to evaluate the performance of the proposed force grasping approach.Note to Practitioners—Robotic grasping often needs to handle novel categories of objects. As a result, frequent retraining of the classification neural network is a pain point, which is tedious and prone to overfitting with only a few samples. In this work, metric learning is introduced for grasping classification where a novel kind of weight imprinting classification is proposed to handle the novel classes by better feature embedding and directly setting the classifier weights without retraining or fine-tuning. Together with the benefits from the variable stiffness feature of the gripper, the proposed vision-based force grasping approach can handle a wide range of objects from fragile to heavy, and the grasping force is controllable from 0.2 N onward to the motor limitation. The controllable grasping force resolution of the proposed vision grasping is better than 0.05 N, the accuracy of the grasping force is evaluated from 0.2 to 12 N, and the evaluated grasping objects are from extremely fragile potato chips and eggshell to heavy flange and metal block.

We aim to tackle the challenging Few-Shot Object Detection (FSOD), where data-scarce categories are presented during the model learning. The failure modes of FasterRCNN in FSOD are investigated, and we find that the performance degradation is mainly due to the classification incapability (false positives) caused by category confusion, which motivates us to address FSOD from a novel aspect of classification refinement. Specifically, we address the intrinsic limitation from the aspects of both architectural enhancement and hard-example mining. We introduce a novel few-shot classification refinement mechanism where a decoupled Few-Shot Classification Network (FSCN) is employed to improve the final classification of a base detector. Moreover, we especially probe a commonly-overlooked but destructive issue of FSOD, i.e., the presence of distractor samples due to the incomplete annotations where images from the base set may contain novel-class objects but remain unlabelled. Retreatment solutions are developed to eliminate the incurred false positives. For FSCN training, the distractor is formulated as a semi-supervised problem, where a distractor utilization loss is proposed to make proper use of it for boosting the data-scarce classes, while a confidence-guided dataset pruning (CGDP) technique is developed to facilitate the few-shot adaptation of base detector. Experiments demonstrate that our proposed framework achieves state-of-the-art FSOD performance on public datasets, e.g., Pascal VOC and MS-COCO.

Dynamic control for robotic automation tasks is traditionally designed and optimized with a model-based approach, and the performance relies heavily upon accurate system modeling. However, modeling the true dynamics of increasingly complex robotic systems is an extremely challenging task and it often renders the automation system to operate in a non-optimal condition. Notably, many industrial robotic applications involve repetitive motions and constantly generate a large amount of motion data under the non-optimal condition. These motion data contain rich information, and therefore an intelligent automation system should be able to learn from these non-optimal motion data to drive the system to operate optimally in a data-driven manner. In this paper, we propose a learning-based controller optimization algorithm for repetitive robotic tasks. To achieve this, a multi-objective cost function is designed to take into consideration both the trajectory tracking accuracy and smoothness, and then a data-driven approach is developed to estimate the gradient and Hessian based on the motion data for optimization without relying on the dynamic model. Experiments based on a magnetically-levitated nanopositioning system are conducted to demonstrate the effectiveness and practical appeals of the proposed algorithm in repetitive robotic automation tasks.

Data-efficient domain adaptation with only a few labelled data is desired for many robotic applications, e.g., in grasping detection, the inference skill learned from a grasping dataset is not universal enough to directly apply on various other daily/industrial applications. This paper presents an approach enabling the easy domain adaptation through a novel grasping detection network with confidence-driven semi-supervised learning, where these two components deeply interact with each other. The proposed grasping detection network specially provides a prediction uncertainty estimation mechanism by leveraging on Feature Pyramid Network (FPN), and the mean-teacher semi-supervised learning utilizes such uncertainty information to emphasizing the consistency loss only for those unlabelled data with high confidence, which we referred it as the confidence-driven mean teacher. This approach largely prevents the student model to learn the incorrect/harmful information from the consistency loss, which speeds up the learning progress and improves the model accuracy. Our results show that the proposed network can achieve high success rate on the Cornell grasping dataset, and for domain adaptation with very limited data, the confidence- driven mean teacher outperforms the original mean teacher and direct training by more than 10% in evaluation loss especially for avoiding the overfitting and model diverging.

Data-driven control methods for modern controller design are becoming popular recently. However, the traditional Proportional-Integral-Derivative (PID) controller is still the most widely used controller to the industrial preference. To tune the parameters of the PID controller, optimal PID tuning approaches such as solving the Riccati equation of the Linear Quadratic Regulator (LQR) provide the optimal solution. The disadvantages of the LQR are that an accurate model of the system is required, and the high-order system must be reduced to the second-order system so that the Riccati equation can be solved. In this paper, a novel data-driven method is proposed to cope with these problems. For the system which is difficult to be identified accurately, the proposed data-driven method can skip the procedure of system identification and tune the parameters of the PID controller directly with the experimental data instead of solving the Riccati equation. This data-driven tuning method also ensures that the parameters of the PID controller for the high-order system are optimized without using the reduced-order model of the system. Simulations are conducted on a tray indexing system with the second-order model and the full-order model demonstrating high applicability and accuracy of the proposed method.

The handling of ultrathin wafers (<;100 μm thickness) is a challenging task since these are among the thinnest and most fragile materials. This paper provides a soft-acting and noncontact gripping technology for ultrathin wafer based on the distributed Bernoulli principle, and also proposes an experimental measurement method for evaluating the performance. A distributed Bernoulli gripper for ultrathin wafers is designed, and the characteristics of the gripper are studied via theoretical analysis and experiments. Three performance indices for evaluating the properties of the soft gripping: deformation, vibration, and stress are presented. Through measurement experiments, the effects of the key operational parameters consisting of air flow rate and gap height on the performance indices are investigated. Based on the experimental data, the appropriate parameters settings are obtained. The comparison to present grippers reveals that the proposed gripping technology is superior in soft gripping thin and fragile materials. This paper provides guidance for implementing the distributed Bernoulli principle in practical applications of soft-acting and noncontact gripping for thin and fragile materials.

The dual-drive H-gantry is widely used for high-speed, high-precision Cartesian motion. Compared with the conventional rigid-linked design, the flexure-linked counterpart is able to prevent the damage of joints for its smaller interaxial coupling force. However, there are still barriers to further push up its precision, such as parametric uncertainties due to the inaccurate dynamical model, the possible induced vibration during high-speed motion, and the decentralized control structure required by industries. To maintain the tracking precision of carriages and minimize the vibration of the end effector, we aim to optimize parameters in decentralized controllers with choices of flexure pieces. We find that such decentralized feedback structure yields some uncontrollable but stabilizable states in the closed-loop system, and no direct solution from solving the algebraic Riccati equation is available in this case. Such structural constraint, together with constraints due to stability requirement and model uncertainties facilitates us to formulate an H2guaranteed cost control problem within a projected convex domain. From here, efficient numerical procedures are developed to obtain the global optimum by iterative linear programming. The real-time experiment validates the optimality and the robustness of the proposed method.

Thermal error in machine tools arises from error in positioning the tool due to thermal distortion of the machine structure. Thermal error can be effectively reduced by implementing a compensation algorithm into the machine control loop. The compensation algorithm is derived from a disturbance model or mathematical representation of the thermal error. In this paper, thermal error in a spindle motor assembly is predicted from temperature measurement through the disturbance model. Model reduction technique is applied to eliminate the temperature sensor, which has least effect on the model output (thermal error). Applying the sensor selection method resulted in a reduced number of temperature sensors (10 to 4) required to achieve the same model prediction accuracy. In the final experiment, the spindle motor is subjected to a 5-h cyclical heat load and the maximum prediction accuracy achieved using the reduced set sensors is 1.5 μm (time averaged accuracy of 0.77 μm).

The disturbance observer (DOB) is widely used in high precision motion control applications as an effective means of rejecting disturbances. DOB design essentially boils down to the design of the Q filter. In traditional DOB design, the Q filter follows a standard form such as the Butterworth or binomial, where the bandwidth is the only tunable parameter. In this paper, we employ a more general form of Q filter, and tune the parameters iteratively using a data-based approach according to some design criterion. Simulation results show that the data-based design achieve improved disturbance rejection in low frequency region while the high frequency noise attenuation and robust stability is not compromised.

Linear direct feed drives are widely used in machine tools, but an abrupt counter force from the secondary part will induce the jerk to the metro frame contacted with the linear motor and cause the vibration of auxiliary devices on it. The jerk-decoupling cartridge (JDC) provides a buffer to reduce such an impact. To systematically take care of both the tracking error and the jerk induced to the metro frame, this paper presents an integrated design approach to determine parameters in the JDC and the position controller of the feed drive. The initial formulated non-convex optimization problem is converted to convex constrained gradient optimization problem and linear step searching problem. Thus, fast convergence of parameters is achieved within first few iterations. Through a series of simulation, the effectiveness of proposed methodology is verified.

Jerk derivative feedforward control is proposed and used in recent literatures. However, such feedforward control scheme is restricted to the minimum-phase system which has constrained static gains in the flexible modes. This paper presents a novel form of feedforward controller for a wider class of flexible motion systems, even to be non-minimum phase, as long as the summation of DC gains of flexible modes is positive. The newly proposed feedforward scheme gives superior performance over existing feedforward schemes. Especially, the tracking error can be almost eliminated, when the plant's lowest frequency anti-resonance is outside the main frequency band of the reference trajectory. In other cases, this novel controller still gives adequate tracking performance, while other feedforward controllers are malfunctioned.

Valve Stiction is a common nonlinear phenomenon in pneumatic control valves and it causes oscillations in the control loops. A model of valve stiction that is easy to implement and accurate is desired for analysis of this phenomenon. Compared with the physical model, the data-driven model does not require excess knowledge on various physical parameters, thus it is widely used in modeling and diagnosis of valve stiction behavior. In this paper, modifications are made to the Two-layer binary tree data-driven model to overcome its shortcomings on handling instantaneous input command on reverse motion. It has simpler logic structure compared with recent proposed XCH model. Accuracy of the revised binary tree model is then tested and validated by ISA control valve standard test.

The scanning accuracy of piezoelectric mechanisms over broadband frequencies is limited due to inherent dynamic hysteresis. This phenomenon has been a key bottleneck to the use of piezoelectric mechanisms in fast and precision scanning applications. This paper presents a systematic model identification and composite control strategy without hysteresis measurement for such applications. First, least squares estimation using harmonic signals is applied to achieve the Preisach density function. Next, the hysteresis output is estimated, such that the non-hysteretic dynamics can be identified. The discrete composite control strategy is proposed with a feedforward-feedback structure. The feedforward controller is the primary component designed for the performance. The secondary proportional-integral (PI) feedback controller is employed to suppress disturbances for robustness. Finally, the identification and composite control strategy is implemented with a dSPACE 1104 board for a real piezoelectric actuator setup. The experimental results indicate that adequate scanning performance can be sustained at a rate higher than the first resonant frequency.

This paper addresses the physical dynamic modeling and simulates the adaptive control of a precision gantry stage. The stage is posed as a three-degree-of-freedom system, based on a patented air-bearing stage provided by the Singapore Institute of Manufacturing Technology. Using this stage as the target model, a mathematical model is then built using the Lagrangian equation. An adaptive control method is formulated for the positioning of the stage, with minimal a priori information assumed of the model. A stability analysis is given for the proposed control scheme. Simulation results are also documented to illustrate the feasibility of the scheme. Future direction is to implement the adaptive controller to the actual gantry stage.