| 摘 要: 强化学习(RL)算法因其强大的环境探索能力,在机器人辅助微创手术自动化探索方面展现了独到的优势。目前已有兼容如dVRK手术机器人模型的微创手术机器人模拟仿真平台SurRoL,可为模拟自主手术提供应用程序编程接口(API)并训练学习策略。为了提高平台中各项仿真手术任务的运行成功率,提出了基于A3C算法原理设计的SO-A3C强化学习策略。相比于仿真平台已有可供训练的其他强化学习策略,SO-A3C策略面对不同难度的手术任务皆表现出优异的性能。训练结果表明所设计的SO-A3C策略相较于现有策略有着更高的综合任务成功率。 |
| 关键词: 机器人辅助微创手术 自主手术 强化学习 A3C算法 |
|
中图分类号:
文献标识码: A
|
|
| Application of SO-A3C reinforcement learning strategy in Autonomous operation of surgical robots |
|
LIU Jianxiong, NIU Guojun
|
(School of Mechanical Engineering, Zhejiang Sc-i Tech University, Hangzhou 310018, China)
liujianxiong202207@163.com; niuguojun@zstu.edu.cn
|
| Abstract: Reinforcement Learning (RL) algorithms have demonstrated unique advantages in automating robotassisted minimally invasive surgery due to their robust environmental exploration capabilities. Currently, the SurRoL simulation platform, which is compatible with the dVRK surgical robot model, provides Application Programming Interfaces (APIs) and enables the training of learning strategies for simulated autonomous surgical procedures. To enhance the task success rate of various simulated surgical operations on this platform, this study proposes an SO-A3C reinforcement learning algorithm based on the A3C(Asynchronous Advantage Actor-Critic)strategy. Compared with other RL strategies available in the simulation platform, the SO-A3C approach exhibits superior performance across surgical tasks of varying complexity levels. Training results demonstrate that the proposed SO-A3C strategy achieves significantly higher overall task success rates compared to existing baseline strategies. |
| Keywords: robo-t assisted minimally invasive surgery autonomous surgery reinforcement learning A3C algorithm |
