Grasp Detection Algorithm Based on Deep Learning

Huiyan Han; Wanning Li; Jiaqi Wang; Liqun Kuang; Xie Han

doi:10.3788/LOP242181

Laser & Optoelectronics Progress, Volume. 62, Issue 12, 1237001(2025)

Grasp Detection Algorithm Based on Deep Learning

Huiyan Han^1,2,3、*, Wanning Li^1,2,3, Jiaqi Wang^1,2,3, Liqun Kuang^1,2,3, and Xie Han^1,2,3

Author Affiliations

¹School of Computer Science and Technology, North University of China, Taiyuan 030051, Shanxi , China

²Shanxi Provincial Key Laboratory of Machine Vision and Virtual Reality, Taiyuan 030051, Shanxi , China

³Shanxi Province's Vision Information Processing and Intelligent Robot Engineering Research Center, Taiyuan 030051, Shanxi , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(20)

Fig. 1. Five-dimensional grasp representation

Download full size

Fig. 2. Overall structure of APNet

Download full size

Fig. 3. AKConv structure

Download full size

Fig. 4. Improved AKConv structure

Download full size

Fig. 5. EMA structure

Download full size

Fig. 6. Residual structure based on PyConv

Download full size

Fig. 7. Visual comparison of grasping detection results of different algorithms on Cornell dataset

Download full size

Fig. 8. Visual comparison of grasping detection results of different algorithms on Jacquard dataset

Download full size

Fig. 9. Visualization of grasping detection results of APNet algorithm on multi-target dataset

Download full size

Fig. 10. Visual comparisons of grasp detection results of GR-ConvNet and APNet algorithms under different occlusion conditions

Download full size

Fig. 11. Visual comparisons of grasping detection results between GR-ConvNet and APNet algorithms under different lighting conditions

Download full size

Fig. 12. Visual comparisons of grab-detection results between GR-ConvNet and APNet algorithms under different camera angles

Download full size

Fig. 13. Robot arm grasping experimental platform

Download full size

Fig. 14. Grasp process examples. (a) Initial state; (b) grasp execution; (c) put; (d) return to initial state

Download full size

Table 1. Hyperparameter sensitivity analysis of APNet algorithm on Jacquard dataset
View table
Table 1. Hyperparameter sensitivity analysis of APNet algorithm on Jacquard dataset
Epoch Learning rate Batch size Accuracy /%
100 0.001 8 95.8
100 0.001 16 95.8
100 0.010 8 93.7
100 0.010 16 93.7

Table 2. Comparative results of different algorithms on Cornell dataset

View table

Table 2. Comparative results of different algorithms on Cornell dataset

Algorithm	Accuracy /%		Time /ms
Algorithm	Image-wise split	Object-wise split	Time /ms
GG-CNN^［10］	73.0	69.0	19
GR-ConvNet^［11］	97.7	96.6	20
CGNet^［12］	97.9	96.7	40
Ref.［13］	98.9	97.1	24
AAGDN^［14］	99.3	98.8	18
SE-RetinaGrasp^［15］	97.0	95.8	23
PTGNet^［16］	98.2	96.9	41
APNet	99.3	98.5	9

Table 3. Comparative results of different algorithms on Jacquard dataset
View table
Table 3. Comparative results of different algorithms on Jacquard dataset
Algorithm Accuracy /% Time /ms
GG-CNN^［10］ 84.3 20
GR-ConvNet^［11］ 94.6 20
Ref.［13］ 95.6 —
AAGDN^［14］ 96.2 —
PTGNet^［16］ 94.8 42
APNet 95.8 10

Table 4. Results of ablation experiments on the Cornell dataset

View table

Table 4. Results of ablation experiments on the Cornell dataset

Network	Accuracy /%		Model size /MB	Time /ms
Network	Image-wise split	Object-wise split	Model size /MB	Time /ms
Baseline network	97.7	96.6	7.60	20
Baseline network+AKConv	98.1	96.9	7.03	16
Baseline network+AKConv+Hardswish	98.3	97.1	7.03	14
Baseline network+AKConv+Hardswish+EMA	98.9	97.8	7.03	14
Baseline network+AKConv+Hardswish+EMA+PyConv	99.3	98.5	5.98	9

Table 5. Results of ablation experiments on the Jacquard dataset

View table

Table 5. Results of ablation experiments on the Jacquard dataset

Network	Accuracy /%	Model size /MB	Time /ms
Baseline network	94.6	7.60	20
Baseline network+AKConv	94.9	7.04	17
Baseline network+AKConv+Hardswish	95.1	7.04	14
Baseline network+AKConv+Hardswish+EMA	95.5	7.04	14
Baseline network+AKConv+Hardswish+EMA+PyConv	95.8	6.01	10

Table 6. Grasp experimental results for statistics
View table
Table 6. Grasp experimental results for statistics
Object Number of successful captures Grasp success rate /%
Pen 48 96
Toy car 45 90
Spoon 44 88
Brick 49 98
Mouse 46 92
Ball 42 84
Bowl 49 98

Tools

Get Citation

Copy Citation Text

Huiyan Han, Wanning Li, Jiaqi Wang, Liqun Kuang, Xie Han. Grasp Detection Algorithm Based on Deep Learning[J]. Laser & Optoelectronics Progress, 2025, 62(12): 1237001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites