在堆叠场景中,用于连续机器人抓握的双阶段抓握检测方法
Jing Zhang1,2, Baoqun Yin1, Yu Zhong2
1Department of Automation, University of Science and Technology of China, Hefei 230027, China.
Mathematical biosciences and engineering : MBE
|March 8, 2024
概括
本研究介绍了对堆叠对象的两相机器人抓取方法,在模拟和现实世界的实验中取得了高的成功率. 这种方法在复杂的堆叠场景中增强了机器人操纵能力.
科学领域:
- 机器人技术 机器人技术 机器人技术
- 计算机视觉 计算机视觉
- 人工智能的人工智能
背景情况:
- 巧妙的抓取对于机器人精细操纵至关重要,但在堆叠场景中具有挑战性.
- 现有的方法在对堆叠的对象进行顺序抓取方面遇到了困难.
研究的目的:
- 提出一种新的两相方法,用于连续的机器人抓取堆叠任务的抓取检测.
- 为了提高机器人掌握在复杂堆叠环境中的准确性和成功率.
主要方法:
- 开发了一个旋转YOLOv3 (R-YOLOv3) 模型,用于在堆叠的场景中检测顶层物体.
- 创建了一个堆叠的场景数据集用于培训和测试R-YOLOv3网络.
- 使用G-ResNet50模型来确定最上层物体的最佳抓取姿势.
主要成果:
- 在康奈尔大学掌握数据集上,R-YOLOv3模型实现了96.60%的平均掌握预测成功率.
- 在现实世界的实验中,机器人在堆叠的场景中实现了95.00%的最大抓取成功率和83.93%的平均处理成功率.
结论:
- 拟议的两阶段方法有效地使机器人能够在复杂的堆叠环境中执行顺序抓取.
- 该方法在涉及堆叠对象的机器人操纵任务中表现出高效率和竞争力.
相关概念视频
01:31Hand Control During Infancy and Toddlerhood
Hand movements during infancy and toddlerhood develop from broad, relatively uncontrolled actions into increasingly precise fine motor skills. This progression depends on nervous system maturation, improved muscle control, and repeated interaction with objects. Early reflexive movements gradually give way to intentional reaching, grasping, object manipulation, and tool use.Early Grasping SkillsNewborns show a strong palmar grasp reflex, automatically closing their fingers when the palm is...
01:29Sensorimotor Stage I: Substages 1 to 3
Piaget’s theory describes cognitive development as a sequence of stages. The first is the sensorimotor stage, which extends from birth to approximately two years. During this period, infants gain knowledge mainly through sensory experiences and motor actions. Over time, behavior progresses from automatic reflexes toward increasingly intentional and goal-oriented actions. The first three substages show the early part of this transition.Substage 1: Reflexive SchemesReflexive schemes span the...