Scholarlink

We present the implementation of a simple and lightweight model for 6D object pose estimation for robotic grasping tasks. Our approach has obtained great results on several industrial scenarios. Lightweight-ness is achieved through the use of neural network backbones inside our architecture. These backbones have great performance for feature extraction, which we exploit to achieve our results. Our approach uses a 6 dimensional space to represent rotation, providing benefits to conventional approaches like Euler angles and quaternions.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Efficient 6D Object Pose Estimation for Robotic Grasping Using Lightweight Neural Network Architectures

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Alejandro Grajeda,
  • David Castro,
  • Jawad Masood

摘要

We present the implementation of a simple and lightweight model for 6D object pose estimation for robotic grasping tasks. Our approach has obtained great results on several industrial scenarios. Lightweight-ness is achieved through the use of neural network backbones inside our architecture. These backbones have great performance for feature extraction, which we exploit to achieve our results. Our approach uses a 6 dimensional space to represent rotation, providing benefits to conventional approaches like Euler angles and quaternions.