Related Experiment Video
Updated: May 31, 2026
09:41
Estimation of Contact Regions Between Hands and Objects During Human Multi-Digit Grasping
Published on: April 21, 2023
AADFNet: An adaptive asymmetric dual-branch fusion network for background-robust grasping
Tao Fan1, Chenyang Liu1, Qiuyang Dai1
1School of Automation, Southeast University, Nanjing, 210096, China; Key Laboratory of Measurement and Control of Complex Systems of Engineering, Ministry of Education, Nanjing, 210096, China.
Summary
We developed AADFNet, a novel network for robust robotic grasping in complex backgrounds. This adaptive dual-branch fusion approach enhances accuracy and efficiency for mobile manipulators.
Area of Science:
- Robotics
- Computer Vision
- Artificial Intelligence
Background:
- Robotic grasping performance degrades in diverse, unpredictable backgrounds.
- Existing RGB-D grasp detection methods lack background robustness on resource-constrained platforms.
Purpose of the Study:
- To propose AADFNet, an Adaptive Asymmetric Dual-branch Fusion Network for background-robust robotic grasping.
- To improve accuracy and efficiency of grasp detection for mobile manipulators in cluttered environments.
Main Methods:
- Developed an Asymmetric Dual-Branch Encoder (ADE) to process RGB and depth data, separating object features from background noise.
- Implemented a Cross-Modal Coordinate Attention (CM-CA) module for deep fusion of RGB and depth information.
- Introduced an Adaptive Multi-scale Feature Decoder (AMFD) for precise grasp localization using dynamic receptive fields.
- Created the GAA synthetic dataset for training and evaluation.
Main Results:
- AADFNet demonstrates competitive performance with a significantly reduced model size.
- Experiments show improved grasp detection accuracy and efficiency in challenging background conditions.
- Real-world tests on a mobile manipulator confirm the system's practicality and effectiveness.
Conclusions:
- AADFNet effectively addresses the challenge of background-robust grasping for mobile manipulators.
- The proposed network architecture offers a practical and efficient solution for real-world robotic grasping tasks.
- The GAA dataset facilitates systematic training and evaluation of grasp detection algorithms.
