Curiosity model policy optimization for robotic manipulator tracking control with input saturation in uncertain environment

• DOI: 10.3389/fnbot.2024.1376215
• Publication Date: 2024-05-01T05:07:47Z
• Authors: Tu Wang; Fujie Wang; Zhongye Xie; Feiyan Qin
• Keywords: Curiosity; Tracking (education)

In uncertain environments with robot input saturation, both model-based reinforcement learning (MBRL) and traditional controllers struggle to perform control tasks optimally. this study, an algorithmic framework of Curiosity Model Policy Optimization (CMPO) is proposed by combining curiosity approach, where tracking errors are reduced via training agents on gains for model-free controllers. To begin with, a metric judging positive negative proposed. Constrained optimization employed update the ratio, which improves efficiency agent training. Next, novelty distance buffer ratio defined reduce bias between environment model. Finally, CMPO simulated baseline MBRL algorithms in robotic designed non-linear rewards. The experimental results illustrate that algorithm achieves superior performance generalization capabilities.