This paper discusses a new testbed developed at the Stanford Aerospace Robotics Laboratory (ARL) to address some of the key issues associated with semi-autonomous construction in a hazardous environment like space. The new testbed consists of a large two-link manipulator carrying two smaller two-link arms. This macro/mini combination was developed to be representative of actual space manipulators, such as the SSRMS/SPDM planned for the Space Station. This new testbed will allow us to investigate several key issues associated with space construction, including teleoperation versus supervised autonomy, dexterous control of a robot with flexibility, and construction with multiple robots. A supervised autonomy approach has several advantages over the traditional teleoperation mode, including operation with time delay, smart control of a redundant manipulator, and improved contact control. To mimic the dynamics found in space manipulators, the main arm was designed to include joint flexibility. The arm operates in 2-D, with the end-point floating on air-bearing. This setup allows cooperation with existing free-flying robots in the ARL. This paper reports the first experiments with the arm which explore the advantages of moving from teleoperation or human-in-the-loop control to the human supervisory or task-level control. A simple task, such as capturing a satellite-like object floating on the table, is attempted first with the human directly driving the end-point and second with the human directing the robot at a task-level. Initial experimental results of these two control approaches are presented and compared.