Stability-Based Sequence Planning for Robotic Dry-Stacking of Natural Stones

The ability to autonomously build strong, dry-stacked structures with in-situ objects would enable new robotic applications in disaster response and remote site preparation. Previous works have demonstrated that heuristics adapted from masonry manuals allow robots to build freestanding dry-stacked structures but have failed to evaluate the strength of the structures. In this paper, we focus on the stability of the dry-stacked structures in both simulation and physical implementation. To estimate the strength of a wall, we physically construct and destructively test a single wall building plan multiple times. We then validate that the simulated strength predicts the strength of the physical walls. We show that evaluating strength throughout the planning process can significantly improve the final stability of structures. Furthermore, we observe that stronger walls are easier to build with a robot because they are more resilient to execution errors. We also investigate the role of execution errors that are due to the limited manipulation skills of a position controlled pinch grip for object placement and show that it is not necessarily a limiting factor to building strong walls.