Biologically-Inspired Control using Pneumatic Actuators

We have been using pneumatic actuators for our ISAC humanoid robot. For the position control of a joint, driven by two artificial muscles, a physical actuator model was designed and used as the basis for a subsidiary torque control [1]. It is known that such actuators contain a high nonlinearity including a hysteresis. Experiments show that the static hysteresis is less important than the dynamic one.

The research focus was on the modification of a physical static model and the extension with a dynamic part. The quality of the model was verified by implementing it as a torque controller and running experiments on a testbed (see Figure 1). The entire arm consists of six joints. In each joint, two artificial muscles are combined to one actuator. This is the same principle human joints are built, where two muscles, called agonist and antagonist, are working against each other to allow movement in both directions and to stabilize the joint. Joachim Schröder, a visiting research student from the University of Karlsruhe in Germany, developed a testbed for this research project in the CRL.

Figure 1. Testbed designed in-house

Artificial Muscles

Artificial muscles were first developed in the 1950’s by the physician Joseph L. McKibben. Thirty years later, Bridgestone Corporation commercialized the muscles called Rubbertuators for industry (Figure 2). Rubbertuators feature hysteresis in their operation due the original hysteresis inherent in the rubber, friction between the rubber and cords and friction between the cords themselves.

Figure 2. ISAC’s Bridgestone Rubbertuators

Structure of artificial muscles movement and the safe interaction with humans because of lower power and stillness than industrial robots. Shadow Robot Company (http://www.shadow.org.uk/)

in London, England began production of artificial muscles for robots in the 1990’s (Figure 3). We replaced most of ISAC’s Soft Arm muscles with Shadow Air Muscles and also used these on the testbed developed in-house. The Air Muscle work principle is the same as Bridgestone Rubbertuators.

Figure 3. Shadow air muscle

Another manufacturer of artificial muscles is the Festo Company. First developed for industrial applications, the Festo muscles were later also used in humanoid robotics at the University of Karlsruhe. Their most famous project based on artificial muscles is the six-legged walking machine Airbug. Several projects running on other Universities show that the interest of artificial muscles for robot applications is still high and will be subject of research in the next years.

Arm Control

Actuator

The entire arm consists of six joints. In each joint, two artificial muscles are combined to one actuator. This is the same principle human joints are built, where two muscles, called agonist and antagonist are working against each other to allow movement in both directions and to stabilize the joint.

Control Architecture

The Soft Arm is not a highly dynamic parts are connected in a serial way, needs to be fast and robust. Such completed with a regular PID controller. To improve speed and quality controller was chosen to build a control loop. The structure shown minimizes errors given to the slow control loop. A dynamic model of to consider masses of inertia and A inner torque control loop with could result in the best control torque or force sensors and the would be required. In this case, about the actuator would be necessary, describe the behavior would have hysteresis and nonlinearity of the pneumatic be controlled easily.

Due to the extra effort in additional costs, wiring and placement of all control feedback is realized with This model must be exactly the muscle actuator, calculating the disadvantage of this procedure is that level cannot be measured and control loop. Therefore, we built an actuator model in-house using the Shadow air muscles (Figure 4).

Figure 4. Pneumatic actuator testbed

Reference

J. Schröder, D. Erol, K. Kawamura and R. Dillman, “Dynamic Pneumatic Actuator model for a model-based Torque Controller”, IEEE International Symposium on Computational Intelligence in Robotics and Automation (CIRA), July 2003, Kobe, Japan, pp.342-347, 2003