Robotic high-precision assembly is a challenging and important technology, as it involves contact-rich operations with a much smaller solution space. This paper aims to propose a strategy to address high-precision assembly problems, and this strategy can quickly adapt to new environments to accomplish new assembly tasks.
This study establishes a deep learning-based force-driven assembly strategy and a vision-driven assembly strategy, and both strategies are able to accomplish the assembly task independently. Then, a Gaussian process-based selection mechanism is proposed to improve the success rate and the adaptability to new tasks. This selection mechanism can balance the strengths and limitations of the force-driven and vision-driven assembly strategies to generate an optimal assembly strategy.
Extensive experiments have been conducted, and the results show that the proposed method can accomplish assembly tasks with a minimum clearance of 0.01 mm and can be easily generalized to new tasks, including variations in assembly clearances, materials, grasping offsets and lighting conditions. Experiment video is available at https://youtu.be/HDdFNupgnwQ.
A Gaussian process-based method is proposed to seamlessly integrate force and vision-driven assembly strategies, enabling rapid adaptation to new assembly tasks. The authors introduce a novel swing strategy that explores the contact state of the robot through controlled swinging motions. Additionally, a self-adaptive strategy is proposed, requiring only a few samples to adapt to new high-precision assembly tasks.
