DSC internal talk in November

Prof. Takamitsu Matsubara gave a lecture in November.

The details are as follows.

====================

Prof. Takamitsu Matsubara (Robot Learning Laboratory)

TITLE:
Learning to Shape by Robotic Grinding: A Cutting-Surface-Aware Model-Based Reinforcement Learning Approach

ABSTRACT：
Object shaping by grinding is a critical industrial process involving material removal by a rotating belt, but automating this with robots requires accurate shape transition models. Learning such models is challenging due to process variability, data requirements, and the irreversible nature of grinding. We proposes a cutting-surface-aware Model-Based Reinforcement Learning (MBRL) method using a geometric cutting model. To address deviations from real data due to removal resistance, two strategies are employed: (1) limited real-world data to learn a residual model, and (2) large-scale simulation data with volume constraints to develop a sim-to-real transferable diffusion-based planner. Simulations and real-world experiments confirm the method’s effectiveness in automating robotic grinding.