Scholarlink

This paper explores the application of Deep Reinforcement Learning (DRL) in human-robot collaborative sorting tasks, emphasizing the robot’s adaptability. DRL enables robots to adjust to dynamic environments, including human presence and unknown objects. Using RGB and depth images processed by DenseNet, the robot recognizes and manipulates objects. The proposed approach estimates grasping positions without prior object knowledge, enhancing adaptability in unstructured environments. A Deep Q-Network (DQN) simulation in CoppeliaSim evaluates execution speed, accuracy, and system robustness, demonstrating improved sorting efficiency and safety in human-robot collaboration.

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

Deep Reinforcement Learning for Adaptive Object Sorting in Collaborative Robotics

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Haris Subašić,
  • Lejla Banjanović-Mehmedović,
  • Selma Subašić,
  • Isak Karabegović,
  • Naser Prljača

摘要

This paper explores the application of Deep Reinforcement Learning (DRL) in human-robot collaborative sorting tasks, emphasizing the robot’s adaptability. DRL enables robots to adjust to dynamic environments, including human presence and unknown objects. Using RGB and depth images processed by DenseNet, the robot recognizes and manipulates objects. The proposed approach estimates grasping positions without prior object knowledge, enhancing adaptability in unstructured environments. A Deep Q-Network (DQN) simulation in CoppeliaSim evaluates execution speed, accuracy, and system robustness, demonstrating improved sorting efficiency and safety in human-robot collaboration.