Control of a flexible-link robotic arm manipulating an unknown dynamic payload.

摘要

When light-weight space-based robots, such as the space shuttle's RMS (remote manipulator system), manipulate massive payloads such as satellites, significant structural bending is induced in the links of the robot. In addition, space-based robots will often manipulate payloads that are not rigid bodies: for example, satellites may contain fuel or have flexible appendages. This dissertation contributes new basic technology that will enable flexible-link manipulators to perform precise end-point control of payloads while simultaneously controlling the unknown internal dynamics of the payloads.;The approach taken here combines high-performance control with an innovative identification algorithm. In addition, to facilitate the development of controllers, a numerically efficient procedure to merge payload and arm dynamic models is presented.;First, controllers incorporating end-point feedback with a known payload are made robust to high-frequency modelling errors, sensor noise, and sensor biases using frequency-weighted linear quadratic gaussian design methods. Using end-point position measurements as the primary sensor, the robot is then capable of actively damping the internal payload dynamics an order of magnitude faster than the natural damping rate--if it knows the payload parameters. However, with this type of controller small parameter variations in the payload can lead to poor performance or instability.;This research has therefore also developed an identification algorithm that updates the end-point controller parameters; this enables the system to achieve high performance when the payload dynamics are not known apriori. By using a nominal control law, the identification problem can be reduced to detecting and identifying eigenvalues of a closed-loop system. To identify these eigenvalues, a generic algorithm capable of determining the eigenvalues of a system in real time has been developed: even the order of the system need not be known in advance. With this approach, moreover, the identification algorithm does not require that the system inject broad-band excitation in order to work accurately.;An experimental robotic system was designed and built to test these emerging control strategies. All of the control strategies have been verified on the experimental robot. Experimental results demonstrate, for the first time, precision end-point control of a very flexible single-link robot arm with unknown dynamic payloads.