Generating robust assembly plans in constrained environments

In the future, teams of robots will construct outposts on Mars and orbital structures in space. Such tasks will require assembly of a large number of components into structures. Automatic generation of assembly sequences is a difficult and well-studied problem in structured factory environments that are specifically engineered for the assembly task at hand, but it is much less understood in less constrained settings. Instead of representing the problem in the space of the many degrees of freedom of the robots and components involved in the assembly, we approach the problem in the space of valid configurations of the structure to be assembled. We use a graph-based framework to describe valid assembly configurations and feasible assembly steps. In addition to reasoning about kinematic feasibility of assembly steps, we consider the quality of potential configurations with respect to actions for the mobile robots. This method automatically repositions the structure in the workspace so that components to be assembled are most approachable. That is, the sequence of assembly and the position of the structure as it is assembled is chosen so as to maximize the area in which mobile robots can operate to perform their tasks. We present simulation results from a simple five-component assembly with and without the constraints of a narrow confined environment. Results show that our method allows over twice as much space available for robots during assembly. In addition, the plans preserve most of their free-space flexibility in tight workspaces where other planning approaches are left with only a few candidate solutions.