Enhancement ofreal‐timegrasp detection by cascaded deep convolutional neural networks

Abstract

Robot grasping technology is a hot spot in robotics research. In relatively fixed industrialized scenarios, using robots to perform grabbing tasks is efficient and lasts a long time. However, in an unstructured environment, the items are diverse, the placement posture is random, and multiple objects are stacked and occluded each other, which makes it difficult for the robot to recognize the target when it is grasped and the grasp method is complicated. Therefore, we propose an accurate, real‐time robot grasp detection method based on convolutional neural networks. A cascaded two‐stage convolutional neural network model with course to fine position and attitude was established. The R‐FCN model was used as the extraction of the candidate frame of the picking position for screening and rough angle estimation, and aiming at the insufficient accuracy of the previous methods in pose detection, an Angle‐Net model is proposed to finely estimate the picking angle. Tests on the Cornell dataset and online robot experiment results show that the method can quickly calculate the optimal gripping point and posture for irregular objects with arbitrary poses and different shapes. The accuracy and real‐time performance of the detection have been improved compared to previous methods.

Gesture recognition based on surface electromyography‐featureimage

“…When the depth of the Conv layer of the CNN is further increased under the premise that the performance of the network is saturated, it will lead to an increase in the number of features that are finally input to the FC layer by the conv layer, then the feature space that can be expressed by the CNN will correspondingly become larger, and the learning ability of the CNN may be strengthened accordingly. The final result is that the calculation cost of the CNN increases and the complexity increases, which is extremely prone to overfitting 64,65 …”

“…The final result is that the calculation cost of the CNN increases and the complexity increases, which is extremely prone to overfitting. 64,65 A CNN model is proposed for the recognition of 52 sEMG gesture movements. As the size of the sEMG images is small, the network layer is deep, which can easily lead to overfitting.…”

Occlusion gesture recognition based on improvedSSD

“…Compared with the two hand gesture recognition rate in the color background and the depth background, the gesture recognition rate in the color background is significantly higher than that in the depth background, which shows that the color background contains more feature information than the depth background. And gesture4 has the lowest recognition rate in color and depth background 61‐63 . Through the analysis of the experimental process, the main reason is that gesture4 has more serious self‐occlusion and less feature information than the other three gestures.…”

“…And gesture4 has the lowest recognition rate in color and depth background. [61][62][63] Through the analysis of the experimental process, the main reason is that gesture4 has more serious self-occlusion and less feature information than the other three gestures.…”

Compound lower limb vibration training rehabilitation robot

“…The Woodway lower limb rehabilitation training robot carries out suspension and standing training for patients. It is cumbersome to wear during rehabilitation training and is not suitable for early bed rest training 31 . At present, the most effective treatment for patients with disabilities and stroke is to use equipment to do repeated limb training and balance exercises.…”