Scholarlink

<p>Handheld assembly technology for dual-arm space robots has become a crucial technology in on-orbit space truss assembly due to its excellent environmental adaptability. However, current methods face several limitations–including fixed roles for both arms, poor adaptability to target structures, and limited algorithmic perspectives–which significantly restrict their practical use. To overcome these issues, this study introduces Tr-PPO: a Transformer-based multi-objective assembly sequence planning method for dual-arm space robots, developed within the Proximal Policy Optimization (PPO) framework. Using a dynamic role assignment mechanism, this approach allows the agent to autonomously generate optimized assembly sequences with switchable dual-arm roles and adaptability to various target truss structures. Specifically, we integrate the Transformer network to improve the autoregressive capabilities of the algorithm, enabling effective capture of long-range dependencies in assembly sequences. Simulation experiments on sequence planning for three typical truss structures demonstrate the effectiveness of our algorithm in planning assembly sequences for different truss configurations, with results demonstrating superior performance across all evaluation metrics.</p>

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

Integrating transformer and PPO for multi-objective assembly sequence planning of dual-arm space robots

  • 期刊论文

  • 发表日期：

    2026年6月23日

查看全文
  • Zixuan Hao,
  • Gang Chen,
  • Yu Liu,
  • Zeyuan Huang

摘要

Handheld assembly technology for dual-arm space robots has become a crucial technology in on-orbit space truss assembly due to its excellent environmental adaptability. However, current methods face several limitations–including fixed roles for both arms, poor adaptability to target structures, and limited algorithmic perspectives–which significantly restrict their practical use. To overcome these issues, this study introduces Tr-PPO: a Transformer-based multi-objective assembly sequence planning method for dual-arm space robots, developed within the Proximal Policy Optimization (PPO) framework. Using a dynamic role assignment mechanism, this approach allows the agent to autonomously generate optimized assembly sequences with switchable dual-arm roles and adaptability to various target truss structures. Specifically, we integrate the Transformer network to improve the autoregressive capabilities of the algorithm, enabling effective capture of long-range dependencies in assembly sequences. Simulation experiments on sequence planning for three typical truss structures demonstrate the effectiveness of our algorithm in planning assembly sequences for different truss configurations, with results demonstrating superior performance across all evaluation metrics.