Physics-Informed Planning of Discrete Lattice Assembly Sequences

Abstract

The assembly of discrete lattice structures using a climbing robot provides a flexible method for the assembly of efficient and large-scale structures. Automating the assembly of any goal structure, and allowing it to be applied to remote locations like the lunar surface, requires the robot to know in which order to place new lattice cells (voxels), and which path to take to get to these locations. For each robot step, which for a small structure can be well over 1000, the partially built structure that supports the robot needs to hold its own weight and the weight of the robot without failing. The inclusion of structural stability constraints in previous assembly sequence planning (ASP) methods for robotically assembled lattice structures is limited, making them suboptimal for structures with complex geometry like unsupported overhangs. Dealing with these structures requires the integration of structural analysis in addition to path planning in an ASP system which can create a plan consisting of assembly sequences and robot paths for the successful assembly of a structure. The plan should be compatible with the experimental setup, which uses magnetically linked 3D-printed PLA voxels of 10 cm in width, assembled by a five-degree-of-freedom robot. To achieve this, this thesis proposes an ASP system based on Ant Colony Optimization (ACO) capable of finding sequences and robot paths that ensure stable substructures throughout the course of the assembly. A path planner is developed to find the steps the robot needs to take to travel between two locations, as well as a FEM solver capable of calculating the internal loads within the lattice structure robot movement. These components are integrated into the Assembly sequence planner. This allows the planner to not only find sequences for simple structures, but also for more complicated structures by dynamically adding scaffolding in locations that need more structural support. The development of a dynamic ant quality function allows for the program to find sequences iteratively while simultaneously optimizing solutions for the minimum number of required scaffolding voxels. By testing this for a variety of different goal structures, one of which was built by the robot using the experimental setup, it is found that the ACO scheme is well suited for the robotic construction of these lattice structures.