CKSRI Seminar Series 2026 “Learning Generalizable Robotic Assembly Policies for Complex Objects"

ABSTRACT:

Assembly is a cornerstone of modern manufacturing and construction, enabling the creation of complex, large-scale products from modular components. Despite its importance, assembly remains a labor-intensive process often requiring hundreds of intricate steps. While robots are increasingly used to automate these tasks, most current systems rely on rigid, pre-programmed procedures that struggle with the "long-horizon" nature of complex assembly. In this talk, we explore the potential of leveraging Artificial Intelligence to revolutionize robotic planning. Drawing inspiration from the success of large-scale pretraining and reinforcement learning (RL) fine-tuning in LLMs, we propose a framework for building robotic assembly policies that are safe, efficient, and highly adaptive to new objects and environments.