Hybrid Motion Planner for a Multi-Armed Robot Performing On-orbit Loco-manipulation Tasks

Large space infrastructure is gaining increasing attention in recent years, including e.g. solar power plants in space or space stations. However, these structures normally are not sent in one piece due to current launch vehicles’ size and load capacity limitations. An alternative is to send individual components and subassemblies and perform the final assembly of structures in orbit, mainly using robotic manipulators that execute simple manipulation tasks. However, fixed manipulator systems have limited reachability; therefore, new approaches are required to endow robotic systems with extended mobility and enhanced versatility.Several European projects have developed initial prototypes of robotic systems that could execute On-orbit services. In the MOSAR project, a 7 degrees of freedom (DoF) Walking Manipulator (WM) robot performed locomotion and manipulation (loco-manipulation) tasks for the assembly of modular satellites. In this project, the robot and the modules of the satellite were equipped with Standard Interconnects (SI), which provide mechanical, power, and data connectivity for the different components and also enable a straightforward way for their manipulation using a suitable SI as a robotic end effector. A ground scenario was developed to evaluate the system under laboratory conditions. To enhance the reachability and mobility of the WM, the project MIRROR, funded by ESA, pursued the development of a Multi-Armed Robot (MAR) capable of executing loco-manipulation tasks to assemble modular space structures. This system consists of three robotic subsystems, two 7-DoF arms attached to a torso with an extra DoF on its base.This paper presents a hybrid motion planner for generating trajectories for loco-manipulation tasks using the MAR system. The developed hybrid planner combines a high-level layer that determines the necessary contact states between the robot and the structure, and a low-level layer that plans the joint trajectories among contiguous contact states. The high-level planner implements task-dependent heuristics and a depth-first search approach to reduce the search time for possible contact states. The low-level layer implements an optimization-based motion planner, in particular the Stochastic Trajectory Optimization for Motion Planning (STOMP). Using an optimization-based approach reduces the random component associated typically with sampling-based motion planners, thus generating as well smoother trajectories. STOMP used as objective function the minimization of joint torques required for executing the different tasks. The performance of the MAR system is evaluated in simulation and on a ground demonstration scenario that simulates the assembly of a modular primary mirror of a space telescope, demonstrating different tasks, including robotic locomotion and manipulation of the mirror parts.