Conventional controllers of industrial robots are decentralized PD joint
controllers. In order to improve trajectory tracking performance, the
control must account for the manipulator dynamic model, i.e. via a
computed-torque technique. An adaptive model following control system
has been developed via hyperstability theory which is aimed at
counteracting robot inaccurate modelling and load variations; the
reference model is usually chosen as a linear decoupled model. This
controller does not require the on-line computation of the dynamic
model, even though partial compensation of model nonlinearities is
possible to decrease the control effort. The resulting control torques
are discontinuous so that typical sliding mode trajectories occur. An
alternative approach has been followed where the control design is based
on a nominal component and a robust, smooth component that depends on
the size of model parameter uncertainties. A microprocessor
implementation of this controller has been proposed. Furthermore, the
extension of the above controllers to the task space to account for
sensory feedback information has been achieved. The problem of designing
model-based controllers for industrial robot sensor-based applications
has been later reconsidered. A geometric fusion technique of
multisensory data has been developed. A parallel decomposition technique
of Newton-Euler recursive inverse dynamics computation has been set up
which allows on-line computation of required torques by means of
Transputers. The potential of control algorithms with (partial)
compensation of manipulator dynamics for improving tracking performance
of industrial robots with typical high gear ratios has been demonstrated
both in simulation and in experiments. A complete dynamic model has been
derived for gear-driven robot manipulators with inclusion of motor
inertia effects. Techniques for identifying the dynamic parameters of
the manipulator model have been implemented on conventional industrial
robots in the PRISMA Lab, as well as at Comau Robotica and Division PMA of KU Leuven.
The problem of finding exciting trajectories has been addressed
in view of the required computational burden. Several model-based
control algorithms have been experimentally tested on the industrial
robot with open control architecture in the laboratory. The design of
robust linear independent joint controllers has been analyzed and a new
scheme has been proposed with an additional acceleration feedback loop
to counteract imperfect modelling and disturbance effects; experimental
tests have been conducted on a high-speed parallel robot.