Experimental Results

The last months Lucy has been assembled and debugged, therefore basic control strategies were implemented. On high level a quasi-static trajectory generation was used to balance the robot and to make it's first steps. On joint level an adapted feedback PI controller was implemented to settle position. The output of this controller is a delta-p signal, which will be added by and subtracted from a chosen mean pressure resulting in two new pressure values for both muscles of a joint. These new values are the set points for the bang-bang pressure controllers.

In order to avoid overshoots the integral term was made adaptive. When pressure differences are too high, the pressure controller won't be able to settle the pressure and thus the integral term will be lowered. In addition different sets for P and I values were used depending whether a leg is support leg or swing leg.

The graphs below show results of the first movements of Lucy. Here Lucy stands up from its rest position with both feet on the ground, lifts up and positions the left leg 0.1m further than the ankle of the stance leg. The same action is than performed by the right leg and next the left leg lifts again and places the feet back together after which the robot finally goes to its rest position.

First experimental steps: angle and pressure values for the left leg

At the left side of the figure angle, desired angle and the difference between them is given for the hip, knee and ankle of the left leg. At the right the pressures in both muscles of each corresponding joint and the control value delta-p is depicted. The definition of the angles can be found in this figure.

Range of motion for the joints

The graphs show clearly the control strategy of keeping the mean pressure constant, which in this case is set at a value of 2.5 bar. One can have an understanding of the antagonistic working principle of the muscles when looking at the pressures for the knee between 10 and 50 seconds. During this time period the left leg switches function from swing leg to stance leg. From 10 to 30 seconds pressure 2, in the flexor muscle, must be higher than pressure 1, in the extensor muscle, to lift the weight of the lower left leg while from 30 to 50 seconds the pressure in the extensor muscle must be higher to hold the weight of the robot. The results of this first control implementation with basic PID techniques show already satisfactory behaviour. Following step will be the implementation of a dynamic control scheme to induce faster and smoother motion for which the local feedback controller has to be redesigned with non-linear control techniques.