Yunsheng Tian MIT CSAIL
时间： 2023-11-06 16:00-2023-11-06 17:30
Assembly planning is the core of product assembly, maintenance, and recycling for modern industrial manufacturing. Despite its importance, current planning processes rely heavily on human effort due to challenges like shape variability and tight physical constraints (motion feasibility, gravitational stability, arm reachability, etc.). To automate this, we first propose a novel method to plan assembly motion efficiently under tight clearances, which leverages the assembly-by-disassembly principle and physics-based simulation to explore a reduced search space. Next, we develop a sequence planning algorithm based on disassembly tree search under all physical constraints and incorporate multiple speed-up techniques. Supported by a few grippers and a flat surface, our system generates physically realistic assembly plans for diverse complex assemblies. We define a large-scale dataset consisting of thousands of physically valid industrial assemblies and show our method’s superior success rates compared to baselines. Finally, we demonstrate the robotic execution of assembly plans both in simulation and through sim-to-real transfer experiments.
Yunsheng Tian is a fifth-year PhD student in the Computational Design & Fabrication Group at MIT CSAIL advised by Prof. Wojciech Matusik. His current research centers on intelligent automation of manufacturing and design by leveraging tools from graphics, robotics, and machine learning. Before coming to MIT, he obtained a bachelor’s degree from Nankai University, advised by Prof. Bo Ren and Prof. Ming-Ming Cheng. He had also worked at Autodesk Research, The University of Hong Kong, and Microsoft Research Asia as a research intern. He had served as a reviewer for SIGGRAPH, CoRL, ICRA, IROS, ICLR, ICML, NeurIPS, etc. Personal website: https://www.yunshengtian.com/