Constrained Path Planning For Robotic Assembly Processes

Kurzfassung

The assembly of large structures, especially modular ones, can be automated using a robot to carry out the process. This can reduce the costs of construction and maintenance of the structures. However, such task brings additional challenges for an automatic planning system. This thesis presents the study and implementation of a hybrid planning system for the demonstration of an in-space robotic assembly of telescopes using segmented mirror tiles. The work was developed within the framework of the EU project Prototype for an Ultra Large Structure Assembly Robot (PULSAR) partially carried out at the German Aeroespace Center (DLR) in Oberpfafenhoffen, Germany. The use of larger telescopes is an increasingly important requirement in the field of astronomy. However, there is a technological limit for producing primary mirrors made of a single rigid piece. Creating a single-piece large reflective surface is normally more expensive than building an array of smaller mirrors designed to act as segments of a primary mirror of the same size. Such problem is especially noticeable for space-based telescopes, where the cargo areas of launch vehicles provide an upper limit for the maximum aperture of the telescope. A possible solution is in-space assembly, making use of a highly autonomous robotic system that directly constructs the structure on orbit. The proposed autonomous assembly planner, takes into consideration different constraints. On the one hand, robot physical constraints such as joint limits, maximum joint torques, and the moments of inertia of the sub-assemblies must be considered. On the other hand, semantic constraints on the types of connections between the tiles must be specified to ensure the mechanical stability and data connectivity within the structure. The final assembly sequence to be executed is selected by minimizing a desired cost function. In order to generate a plan that considers these physical constraints, a hybrid assembly planner with two layers is defined. The global layer uses graph search strategies to explore a graph representation of possible assembly sequences, ensuring that the potential solutions fulfill all the semantic constraints describing the allowed connections between standard interfaces. The local layer uses a constrained path planning algorithm combined with a dynamic simulation to ensure that the transitions between states (representing sub-assemblies) fulfill the required physical constraints. The planner is integrated into a physical ground-based demonstrator for autonomous robotic assembly of mirror tiles. Simulations demonstrate that our approach is still valid for larger telescope structures. In order to illustrate the flexibility of the planner, different physical constraints and optimization goals were implemented and verified.