Assembly Planning for Large Space Structures using Robotic Manipulation

Abstract

There are many ways to explore the vastness of space, and since Galileo, humankind has used telescopes for it. However, the observation of the space is affected by several sources of error, being the influence of the atmosphere a critical one. In order to eliminate this problem, space telescopes are taking a predominant role. However, the necessary systems are becoming larger and larger and can no longer be sent to space in a single launch. In-Space Assembly (ISA) is becoming a key factor in space exploration, as it offers the option to autonomously build large structures in space. The study case of this thesis is centered around how identical Segmented Mirror Tile (SMT)’s elements have to be assembled, with a robot, into a large space telescope. The initial inspiration for this thesis came from the PULSAR project. It was noticed that during the assembly of the mirror, the robot movements had an excessive influence on the pose of the satellite base. Therefore, a hybrid planner was developed to minimize the rotation of the base. This hybrid planner consists of two elements: a physical layer and a logical layer. The physical layer deals with the dynamics of the system, specifically with the influences of the robot motion on the satellite base. Its dynamic computations are based on the conservation of momentum of the system. Also, the physical layer plans and optimizes the trajectories used to place the individual SMT’s. Stochastic Trajectory Optimization for Motion Planning (STOMP) was used to generate and optimize these individual trajectories. The logical layer uses a disassembly-for-assembly strategy to transform the telescope mirror into a graph. Each node represents the current construction step of the telescope, while the edges represent the task of placing a new SMT into the telescope. Each edge has information on the effects of the single building step on the base disturbance, these effects are calculated by the physical layer. In the last part of this thesis, improvements to the hybrid planner are shown. An approximation based on the conservation of momentum is presented. In this approximation, the assembly parts are moved linearly with a simplified trajectory by including intermediate positions. This speeds up the calculation of a cost by a factor of 800. This rough estimate can be used to prune the branches in the logical layer for a faster computation of a feasible solution. Such a solution can be later refined with the original hybrid planner. All elements of the system have been tested either on real hardware or in simulation.