Institut für
Robotik und Prozessinformatik

Deutsch   English

Zur Zeit leider nur in Englisch vorhanden...

Model-Based Control of Robot Manipulators (Ongoing Project)

Project Description

Robots may be regarded as multi-body system, whose outputs are nonlinearly coupled. It is alluring to perform a decoupling of such systems, which theoretically is possible. Commonly, the approaches are based on the inverse dynamic model of the manipulator. As model inaccuracies and other disturbances exist the practical usage is constricted. As the determination of the inverse robot dynamic is more easy than the determination of the direct model, its usage for liberalization becomes facile. But the above-mentioned problems still exist. To compute current joint torques, the state variables position, velocity, and acceleration are required. The two latter quantities cannot be measured directly. Obviously, these quantities can be derived numerically, but it leads to significant disturbances. That makes the usage rather difficult. A new control structure for position control has been analyzed at our institute. The concept is based on the forward model and uses the advantage, that the forward model has integral properties. This means, the state variables can be evaluated much more accurately and without any loss of stability. In this method, the current torques are pre-simulated in an independent loop. That results in a two-loop control structure, which can be seen below.

Block diagram of the control approach.

The controller R controls the model of the plant and the controller R corrects the model inaccuracies of the simulated loop and handles other disturbances. In literature, this type of control structure is called Model Following Control (MFC).

The suggested control structure has been tested on EDDA (Experimental Direct Drive Arm), a two-joint robot arm with direct drives. After the derivation of the EDDA-dynamics, a simulation of the robot has been implemented.A Flash simulation of the experiment can be found here.
. Subsequently, the robot was driven by a single loop PID controller. The results are presented in the following figure. 

Results for PID control.

It shows the step response of the first (left plot) and the second (right plot) joint. The desired values are labeled by q, respectively, and the current joint positions are labeled q. The effect of coupling clearly can be seen. The extreme step of 90 degrees in the first joint generates disturbances in the second joint. The result is an unstable behavior.
The experiment has been repeated for the MFC system. The corresponding results are shown below.

Results for MF control

The additional curve (q) in the plot presents the behavior of the simulated loop. As can be seen, the structure is able to handle the big angle step. The first joint does not oscillate and the second joint remains stable with reduced oscillations.

Further information may be found in our Publications. In case of any further questions please contact Rafael Osypiuk.

It took 0.28s to generate this page.