Cognitive Robotics最新文献

Zero shot learning (ZSL) is aim to identify objects whose label is unavailable during training. This learning paradigm makes classifier has the ability to distinguish unseen class. The traditional ZSL method only focuses on the image recognition problems that the objects only appear in the central part of images. But real-world applications are far from ideal, which images can contain various objects. Zero shot detection (ZSD) is proposed to simultaneously localizing and recognizing unseen objects belongs to novel categories. We propose a detailed survey about zero shot detection in this paper. First, we summarize the background of zero shot detection and give the definition of zero shot detection. Second, based on the combination of traditional detection framework and zero shot learning methods, we categorize existing zero shot detection methods into two different classes, and the representative methods under each category are introduced. Third, we discuss some possible application scenario of zero shot detection and we propose some future research directions of zero-shot detection.

The increase of video content and video resolution drive more exploration of video compression techniques recently. Meanwhile, learning-based video compression is receiving much attention over the past few years because of its content adaptivity and parallelable computation. Although several promising reports were introduced, there is no breakthrough work that can further go out of the research area. In this work, we provide an up-to-date overview of learning-based video compression research and its milestones. In particular, the research idea of recent works on learning-based modules for conventional codec adaption and the learning-based end-to-end video compression are reported along with their advantages and disadvantages. According to the review, compare to the current video compression standard like HEVC or VVC, from 3% to 12% BD-rate reduction have been achieved with integrated approaches while outperformed results on perceptual quality and structure similarity were reported for end-to-end approaches. Furthermore, the future research suggestion is provided based on the current obstacles. We conclude that, for a long-term benefit, the computation complexity is the major problem that needed to be solved, especially on the decoder-end. Whereas the rate-dependent and generative designs are optimistic to provide a more low-complex efficient learning-based codec.

Brain-computer interface (BCI) based on electroencephalogram (EEG) signals can provide a way for human to communicate with the outside world. This approach is independent of the body's peripheral nerves and muscle tissue. The brain-controlled robot is a new technology based on the brain-computer interface technology and the robot control technology. This technology allows the human brain to control a robot to perform a series of actions. The processing of EEG signals plays a vital role in the technology of brain-controlled robots. In this paper, the methods of EEG signal processing in recent years are summarized. In order to better develop the EEG signal processing methods in brain-controlled robots, this paper elaborate on three parts: EEG signal pre-processing, feature extraction and feature classification. At the same time, the correlation analysis methods and research contents are introduced. The advantages and disadvantages of these methods are analyzed and compared in this paper. Finally, this article looks forward to the EEG signal processing methods in the process of brain-controlled robots.

In order to overcome the problem that in a mixed-critical system, once the critical level of the system changes, lower-critical tasks may be abandoned in order to ensure the schedulability of higher-critical tasks. A semi-partitioned scheduling algorithm SPBRC, which is based on a homogeneous multi-processor mixed-criticality platform and integrates the advantages and disadvantages of global scheduling and partitioned scheduling is proposed. First-fit and worst-fit bin-packing algorithms are firstly used in this method to sort high and low critical tasks separately, all high critical tasks as fixed task allocation in different processors in turns, and then distribute the lower-critical tasks. When the criticality of processor changes, lower-cirtical tasks will be allowed to migrate to the processor that is paired with the processor and is in low-critical mode, rather than abandoned. Thus, the overall performance of the system is improved. The simulation experiment verifies the effectiveness of this method in reducing the task loss rate and job loss rate.

Due to the integration of information and internet, the power network has facing more and more uncertain risks of malicious attacks. In response to this problem, we studied it from following four aspects. First of all, multiple cyber-attacks (Denial-of-service, information disclosure, replay attack and deception attack) are analyzed from each operating mechanism. Then, the subsystems are concluded to be a generic modeling frame with considering different type cyber-attack. Secondly, secure defense scenarios are proposed according to each kind cyber-attacks based on the mechanism details. Thirdly, security control conditions are derived by utilizing control theory of stability. Finally, IEEE-14 and IEEE-39 system are used as typical cases to illustrate and analyze the impact of dynamic load altering attack on some of these nodes.

Visual servo is a method for robot arm motion control. It is controlled by the end effector velocity that is the result in calculating the internal Jacobi matrix and vector of feature error. In general, automatic robotic task require high quality sensor that can measure a 3-dimential distance, and do calibration in order to suit the sensor frame and robot frame in Euclidean space. In this paper, we only use RGB camera as the data collection, which not requiring the calibration in sensor frame. Thus, our method is simpler than any other automatic motion methods. Meanwhile, the proposed characteristics based visual servo method has varying the hyper parameter, and show the effectiveness for indicating the precision of pose error both simulation and actual environments.

How to realize data transmission between Intranet and extranet devices in physical isolation is an important problem. QR code can be used to identify text, image information, can be used as a carrier of information exchange. In this paper, QR code technology is used to propose a data transmission system between the Internet and the Internet under the physical isolation state based on visual recognition, which makes the information transmitted only through visible light, enhances the confidentiality, and can meet the daily office needs. At the sending end, the division of the transferred file is completed and multiple corresponding QR codes are generated. After the receiving end scans and recognizes the QR code, it finally forms a complete file identical to the original file. Then, the multi-QR code picture transmission was upgraded to the QR code stream transmission mode, and the transmission efficiency was improved by more than 10 times compared with similar work. In face of the real needs, we have added the function of retrieving pictures or database information from another computer and sending it back to the sending computer according to the instructions of the sending end. The paper also carries on the test analysis to each work.

In this paper, the dynamic engagement characteristics of wet clutch are simulated by finite element method. In the fluid friction, the average Reynolds equation is amended and dimensionless parameters are involved, which is applied to calculate the viscous torque. In the boundary friction, a surface elastic contact model is established to calculate rough contact torque. In the mixed friction, total torque consists of viscus torque and rough contact torque. Experimental comparisons between the simulations and the SAE#2 bench tests are provide to verify the validity of the proposed method, the engagement time errors, the output torques maximum errors and the output torques average errors are utmost 4.86%, 3.87% and 0.73% respectively. The proposed method can be used to guide the design of wet clutches in early stages of product development.

This paper proposes a novel algorithm combining object detection and potential field algorithm for autonomous operation of SCARA arm. The start, obstacles, and goal states are located and detected through the RetinaNet Model. The model uses standard pre-trained weights as checkpoints which is trained with images from the working environment of the SCARA arm. The potential field algorithm then plans a suitable path from start to goal state avoiding obstacle state based on results from the object detection model. The algorithm is tested with a real prototype with promising results.

Big data-driven deep learning methods have been widely used in image or video segmentation. However, in practical applications, training a deep learning model requires a large amount of labeled data, which is difficult to achieve. Meta-learning, as one of the most promising research areas in the field of artificial intelligence, is believed to be a key tool for approaching artificial general intelligence. Compared with the traditional deep learning algorithm, meta-learning can update the learning task quickly and complete the corresponding learning with less data. To the best of our knowledge, there exist few researches in the meta-learning-based visual segmentation. To this end, this paper summarizes the algorithms and current situation of image or video segmentation technologies based on meta-learning and point out the future trends of meta-learning. Meta-learning has the characteristics of segmentation that based on semi-supervised or unsupervised learning, all the recent novel methods are summarized in this paper. The principle, advantages and disadvantages of each algorithms are also compared and analyzed.