Fig. 1. Flow chart of network model

Fig. 2. Schematic of grasp posture

Fig. 3. Instance normalization layer

Fig. 4. Feature pyramid network for semantic segmentation

Fig. 5. High precision lightweight grasping network based on fully convolutional network

Fig. 6. Examples of instance normalization. (a) Convolutional layer; (b) residual block; (c) transpose convolutional layer

Fig. 7. Comparison of stability among various methods

Fig. 8. Qualitative detection results of image grasping recognition based on Cornell grasping dataset. (a) Object partially in figure; (b) translucent object; (c) reflective object; (d) bifurcated object; (e) irregularly shaped object

Fig. 9. Real-time detection for single-target grasp of unknown object detection. (a) Reflective object; (b) irregular object partially in figure; (c) transparent object