Abstract:
An exciting frontier in robotic manipulation is the use of multiple arms at once. However, planning concurrent motions is a challenging task using current methods. A major obstacle is the high-dimensional state space of this planning problem, which renders many traditional motion planning algorithms impractical. This opens the door for alternatives to the common practice of relying on sampling-based techniques.

In this talk, I will present our recent exploration of Multi-Agent Path-Finding (MAPF) algorithms for multi-arm planning. Specifically, I’ll focus on Conflict-Based Search (CBS), which has shown great promise in discrete 2D domains. However, directly applying CBS to multi-arm planning proves inefficient, as its conservative conflict resolution and the high dimensionality of robot-arm configurations slow exploration.

I will cover two approaches to dealing with these challenges. First, by generalizing CBS to admit new types of conflict-resolution constraints that are practically effective for robotic arms but would normally render CBS incomplete. Second, by accelerating CBS algorithms using their repetitive and incremental nature. We’ll see that both methods preserve completeness and bounded sub-optimality guarantees, and watch some demonstrations of simulated and real-world experiments.

Committee:
Maxim Likhachev (co-advisor)
Jiaoyang Li (co-advisor)
Jeff Ichnowski
Muhammad Suhail Saleem