Regularizing Model Predictive Control for pixel-based long-horizon tasks

Open Webpage

Yao-Hui Li, Feng Zhang, Qiang Hua, Chun-Ru Dong

Published: 01 Sept 2025, Last Modified: 06 Nov 2025Applied Soft ComputingEveryoneRevisionsCC BY-SA 4.0
Abstract: Highlights•A sample-efficient RMPC algorithm is proposed for pixel-based long-horizon tasks.•A regularization is incorporated into value function estimation and model learning.•RMPC achieves a 20.88% improvement and a 56.39% decrease in standard deviation.