Pneumatic actuators are more difficult to control than electric motors because of nonlinear effects like compressibility. Many researchers have examined adaptive and nonlinear techniques for producing accurate position control of pneumatic actuators [Binnard 95]. To minimize parasitic effects, these experiments use large components which are inappropriate for a mobile robot.
Despite the apparent difficulty in controlling a nonlinear system with poorly behaved components, Boadicea achieves acceptable force control. Although position control is more difficult, a simple system was sufficient to demonstrate Boadicea's mechanical capabilities.
Boadicea was designed to operate primarily under force control. Pressure sensors are mounted directly on the actuators, providing accurate measurements of leg loading. In contrast, Boadicea's position sensors measure joint angles, which must be converted to actuator position by a kinematic calculation that limits accuracy.
For position control, Boadicea uses a proportional-integral (P-I) control law. The advantage of an integrator is that it provides high gain at low frequencies and attenuates higher frequency noise. The biggest disadvantage the P-I control law is integral windup. Also, the servo response is slow because the valve does not open until after an error is established. To solve these shortcomings, Boadicea uses a feed-forward control law for a short time after a change in goal position. When the goal position changes, the P-I feedback controller is disabled, and the feed-forward controller operates open loop while the leg is moving.
The feed-forward controller provides several advantages. First, because the integrator is disabled during the motion, it does not build up a large error. Second, the response time of the system is improved because the valve opens full bore without waiting for an error to accumulate. A third benefit is that the feedback controller only operates on smaller errors, so the proportional gain, Kp, can be increased without causing instability.
The performance of Boadicea's servo system is certainly not ideal. There are many sources of noise, inaccuracy, and nonlinearity in the system. Despite these difficulties, though, I was able to develop a controller without the computational complexity of an adaptive, fuzzy logic, or neural net controller. The legs do not always move exactly as commanded, but the control system behaves well enough for the robot to demonstrate its capabilities.
This page is still under construction.