Scholarlink

This research explores the integration of reinforcement learning algorithms within a complex environment developed in Unity. The core of this study is not only to train a UR3e robot to complete a complex task, but also to adapt the behavior to the human presence, in order to dynamically adjust its trajectories in real-time. Avoiding humans while being able to do a job such as taking an object, deciding where to leave it and then getting back to the original position, represent progress in human-robot cooperation. By using a virtual setup, real human-robot interactions will be simulated, allowing for a controlled yet realistic examination of the resultant robot’s adaptive behaviors. The implementation of reinforcement learning algorithms enables the UR3e to learn from the environment and modify its actions based on direct human interaction, thus achieving smoother and safer movements. These findings indicate that reinforcement learning can significantly improve human-robot interaction by enabling robots to anticipate and effectively navigate around humans, offering promising results for future applications in various fields, including manufacturing, healthcare, and personal assistance. This study not only demonstrates the feasibility of using reinforcement learning in complex interactive settings but also opens new avenues for enhancing the intuitiveness and safety of human-robot collaborations.

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

Reinforcement Learning for Dynamic Trajectory Adjustment in Human-Robot Interaction Within Virtual Simulations

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Asier Gonzalez-Santocildes,
  • Juan-Ignacio Vazquez,
  • Andoni Eguiluz,
  • P. Garcia Bringas

摘要

This research explores the integration of reinforcement learning algorithms within a complex environment developed in Unity. The core of this study is not only to train a UR3e robot to complete a complex task, but also to adapt the behavior to the human presence, in order to dynamically adjust its trajectories in real-time. Avoiding humans while being able to do a job such as taking an object, deciding where to leave it and then getting back to the original position, represent progress in human-robot cooperation. By using a virtual setup, real human-robot interactions will be simulated, allowing for a controlled yet realistic examination of the resultant robot’s adaptive behaviors. The implementation of reinforcement learning algorithms enables the UR3e to learn from the environment and modify its actions based on direct human interaction, thus achieving smoother and safer movements. These findings indicate that reinforcement learning can significantly improve human-robot interaction by enabling robots to anticipate and effectively navigate around humans, offering promising results for future applications in various fields, including manufacturing, healthcare, and personal assistance. This study not only demonstrates the feasibility of using reinforcement learning in complex interactive settings but also opens new avenues for enhancing the intuitiveness and safety of human-robot collaborations.