Influencing
Robot-Control Performance
Through Data
Tuning
Department of
Mechanical Engineering
University of British
Columbia,
Vancouver, B.C.
Canada
Typical
industrial robot controllers are closed and therefore the development and
technology transfer of control algorithms is inhibited without access to
proprietory systems. Commercially
available industrial controllers suitable for controlling systems at the
frequencies necessary for robots are typically limited in the types of
algorithms that can be implemented, and industrial control engineers prefer
algorithms that that are easily defined and have well defined real time
computation limits. Therefore, in
practice, alternative strategies for improving robot performance are
desirable. This talk focuses on
improving the incoming data to the robot controller. That is, tuning the data to the controller-plant system
rather than tuning the controller to the data.
In
this talk two examples where improvement of the data signal to the controller
are used to improve the control performance of the system. In the first case, an online,
near-optimal, trajectory-planning algorithm which limits jerk is presented. Experimental results demonstrate that
using the same industrial controller a much improved level of tracking can be
achieved.
In
the second example, an approach for improving the data stream through a
heterogeneous sensor-fusion approach is discussed. The approach demonstrates how control can be maintained even
in the presence of degraded
or dazzled sensor data.
3:30 – 4:30
p.m.