Flexible Manipulators | Aerospace Robotics Lab

Feedback Control of Very Flexible Manipulators

Achieving high-performance control of very flexible structures is a challenging task, but one that is critical to the success of many important applications. Fundamental issues that make this task demanding include:

How to control a system where the actuators and sensors are separated by a flexible structure
How to model a complex flexible structure
How to deal with unknown system parameters
How to control end-point impedance to allow contact with the environment

The ARL has a vision of extending and enhancing the utility of flexible structures in ways that are pertinent to all real-world applications. Toward this vision, the ARL has an on-going research program focused on making significant basic advancements in each of these fundamental issues. All of our related research is guided by one common goal: developing a fundamental understanding of how to achieve high-performance control of very flexible structures. Continuing research is critical to the successful overall control of very flexible structures. A most important point about research within the ARL is that all of the research is verified experimentally. This emphasis allows us to directly address real-world challenges and to make fundamental contributions that could not otherwise even be considered. The latest experimental platform is the flexible joint macro/mini manipulator.

Key Technologies

Use of end-point sensing in a feedback control system
Techniques for modelling of complex flexible structures
On-line adaptive techniques.

Experimental Platforms

A single-link very flexible manipulator has been used for many different experiments. This manipulator is driven by a single direct-current motor at the ``hub'', and the sensors include a potentiometer at the hub and a vision system which tracks an array of infrared LED's at the end-point. This system has a payload at the tip that can be modified for use in evaluating adaptive algorithms.
A two-link very flexible manipulator has also been developed for experimental demonstrations. This planar manipulator is suspended on an air bearing, and actuated by direct-current motors at the ``shoulder'' and the ``elbow''. Analog sensors provide angular position measurements at the shoulder and elbow, and a vision system is used to track the end-point. This system has been modified to include a mini-manipulator at the end of the macro-manipulator.
A four-disk system has been used for similar experimental research. The four disks are connected by torsional springs through the center of each disk, so the rotational motion is coupled. This platform is representative of a non-colocated system, where one of the disks is actuated and the position of another is sensed for feedback.

Contributions to Date

In 1982, the ARL was the site of the first-ever experimental demonstration of quick, precise end-point control of a very flexible structure. This landmark research by Eric Schmitz has been an invaluable reference for all other research that has followed in this exciting field.
The control of a very flexible two-link structure is a complex problem, extending the challenges addressed by Schmitz to include non-linear geometric relationships. Once again, the ARL was the site of the first-ever experimental demonstration of quick, precise end-point control of a very flexible two-link structure
One key to the successful end-point control of very flexible structures is a model that accurately describes the dynamics of the structure. However, this is a very difficult task for a system as complex as a flexible structure. The ARL has demonstrated particular success at creating such a system model through the use of physical insight and understanding of the plant as can be gained only through interaction with an experimental system

A next logical concern in the control of flexible structures is to consider the dependence of a closed-loop controller on the system model. Research in the ARL has methodically addressed the problems related to having various unknown system parameters, and the resulting effects which that has on closed-loop system performance.

Dan Rosenthal studied the sensitivity of non-colocated system models for a representative class of flexible structures to uncertainty in various physical parameters, and performed a comparison of robust control design techniques. The results of this work were demonstrated experimentally using a rotary four-disk-and-torsional-spring system
Michael Sidman used this same rotary four-disk system to demonstrate high-performance adaptive control in the face of a destabilizing change in disk inertia. This work led to the development of a ``bumpless transfer'' technique that allows on-line swapping of feedback controllers without unwanted excitation of the plant dynamics
Dan Rovner explored the sensitivity to an unknown tip mass. His experimental work led to fundamental advances in adaptive control theory, including important new insight into the basic interaction between adaptation and the realities of experimental applications, such as sensor noise.
Lawrence Alder continued research in this area by studying the sensitivity to an unknown dynamic payload. This work incorporated a novel signal-processing technique in a first-ever experimental application, and demonstrated, for the first time, the ability to adapt to a payload that had unknown dynamics of its own
Due to the importance of having an accurate system model, the ARL has focused a research program on a systematic means to model arbitrary flexible structures. This work, begun by William Ballhaus, employs a global understanding of the system to coordinate the connection of modular subsystem models in a manner that leads directly to the design of a high-performance end-point control system. The technique developed has been demonstrated experimentally on a two-link very flexible structure.

Research in Progress

Steven Ims is currently extending Alder's work to the problem of controlling flexible articulated structures manipulating payloads that themselves contain several bodies having unmodelled dynamics.
H.D. Stevens is currently extending Ballhaus's work to the problem of comming into contact with the environment and regulating tip force for the multiple-link flexible manipulator with a minimanupulator. With the development of an impedance control scheme, and a quantification of the limits of the impedance possible, this class of manipulators can be used just like rigid manipulators are used currently.