The goal of this section is to provide the major insights gained from the experimental simulation (section 2.8) conducted within the scope of this work. It is natural, that user input commands may deviate each running cycle. This might lead to unstable running conditions, such as increasing Apex height changes. Human input commands can be trained based on sensimotor learning tasks, in order to feed the haptic device controller more precisely.

According to the results presented in the previous chapter 6 (Results), changing stiffness parameters has a more successful influence when it comes to achieving long running distances. Maximum running distance achieved with external force application was about 10 :, whereas stiffness manipulation scenarios triple this value, even for slow starts (i.e. 1R = 2.5 :/V). Slow starts are directly related with velocity values under stable running conditions and are characterized by relatively small start velocities. In this particular case, dividing the initial horizontal velocity 1R by two, still leads to remarkably high running distances (> 35 :) achieved by the user. By applying the human-in-the-loop approach, stable running with different initial parameters is possible, since the user is controlling the haptic device robot properly. Within the following trials, where spring stiffness is manipulated by the user (1R = 3.0 :/V and 1R = 4.0 :/V), results are more consistent and improving. Taking the setup with an initial horizontal velocity 1R = 3.0 :/V as an example, in three out of five trials a running distance of 25 : has been reached. In these three trials, the SLIP model falls down while moving forwards, unlike in some scenarios where external force has been applied haptically.

When applying forces instead of manipulating the spring stiffness, the angle of attack (P/ = 70°) is the same at the beginning of each touchdown. After the first ground contact has been computed, the applied force is divided by trigonometric calculations. Instead of using a constant value for the angle of attack P/, a possible extension to the running system might be to implement the flight dynamics of the SLIP model, where the leg orientation is taken into consideration as well (8 Future Work).
\ As investigated in the result section (6 Results), increasing Apex height changes lead to unstable running conditions. This makes sense, when using static parameters that don’t change over time during the simulation process. It can be stated that increasing initial velocity minimizes CoM fluctuations. This applies until the optimal start velocity (approximately 5 :/V) has been reached. If this value is exceeded, further investigations need to be made in order to answer upcoming question concerning the CoM trajectory.

An interesting question to be answered within the scope of this work is, whether manipulating spring stiffness or applying an external force to the point mass during stance phase leads to a more stable running pattern. As shown above, changing the spring stiffness in real-time shows better results with respect to the achieved running distance. This is due to the fact, that changing spring stiffness simulates the biomuscular behaviour better with respect to muscle force production and compensation. Mechanically speaking, muscle force production refers to leg actuation while compensation refers to damping during stance phase. When actuating the virtual representation of the human leg, energy is introduced into the system and the prismatic spring leg gets stiffer. Based on these assumptions, inserting energy into the system during late stance may lead to greater Apex heights depending on the take-off angle.

A stable running pattern can be achieved under the systematic influence of parameter manipulations. Once the locomotion pattern of the SLIP model is regarded as stable, parameter changes may be repeated the same way during each running cycle. The first touchdown of the SLIP model poses a major challenge. Ideally, Apex heights changes differ less while the simulation is executed. An example is given in Fig. 24, where the purple line shows the greatest impact (i.e. ground reaction forces) while the first step is made. The SLIP model seems to compensate the impact of the first touchdown successfully, considering CoM trajectory in U-direction in lower subplot, at the beginning of the simulation, afterwards it seems, that the running model returns back to unstable state.

Due to LabView-intern operational issues with formula nodes (script nodes for mathematical expressions and calculations) used in the SLIP simulation developed within the scope of this work, certain changes needed to be made. The sampling rate was set to 1 :V, which caused the system to crash at the beginning of the simulation. By setting the sampling rate to 2 :V, the system simulation was running fine for a sufficient amount of time. Luckily, the trajectory of the SLIP model did not vary for the same start parameter set, after this change has been made. Changing the sampling rate may affect the system, if the simulation time increases drastically.

Beside the upper mentioned case, no further issues occurred during the experiments.

indexmenu_n_70