Abstract:
Robotic assembly is a critical category of tasks in modern manufacturing. Compared to a single-robot workstation, a multi-robot system offers several advantages: 1) it expands the system’s workspace, 2) improves task efficiency, and, more importantly, 3) enables robots to achieve significantly more complex and dexterous tasks, such as cooperative assembly. However, effectively coordinating the tasks and motions of multiple robots for real-world assembly tasks is challenging due to issues such as system uncertainty, task efficiency, algorithm scalability, and safety concerns.

To address these challenges, I will present our proposed multi-level reasoning framework designed to safely coordinate multiple robot arms and achieve efficient, scalable, and robust assembly. Specifically, I will first introduce how we use a custom physics reasoning tool for physics-aware assembly sequence planning and task reasoning. Then, I will discuss our proposed action reasoning approach and highlight how we can post-process a sequential task and motion plan to enable robust asynchronous execution under uncertainty. I will focus on how we generalize the Temporal Plan Graph (TPG) execution framework to multi-arm coordination, which allows robots to incorporate task-specific manipulation skills, react to failures and anomalies in the assembly process, and adjust to execution delays asynchronously. Experimental results demonstrate that using TPG post-processing can significantly speed up the execution time of many long-horizon LEGO assembly tasks by 48% compared to sequential planning and 36% compared to synchronous planning on average. Finally, I will showcase how our framework is deployed on a dual-arm assembly system to construct complex, customized, and even generative LEGO designs.

Committee:
Jiaoyang Li (Chair)
Max Likhachev
Changliu Liu
Muhammad Suhail Saleem