2 results
Filters
Settings
Now showing1 - 2 of 2
- Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , Combined feedforward and feedback dynamic trajectory tracking control of a soft robotSoft robots are an emerging and fast-growing field of research with applications in various areas. With the rise of new soft robotic applications, control requirements increase. Thus, control of dynamic motion is becoming more and more important. The main challenges in controlling soft robots stem from their large, continuous deformations, which lead to many degrees of freedom, as well as the difficulty of sensor integration, which leads to a limited availability of control feedback. Additionally, unknown material properties and comparatively large manufacturing inaccuracies hinder reliable modeling of soft robots. So far, in soft robotics mainly so-called quasi-static control approaches, which neglect the dynamics of the system, are used. However, modern soft robotic applications demand increasingly faster and more accurate movements. In these cases, quasi-static controllers are insufficient, and controllers which account for the robots dynamics are needed.In this work different feedforward and feedback control approaches for dynamic trajectory tracking control of soft robots are implemented and examined. A neural-network-based quasi-static feedforward controller, a dynamic feedforward controller based on the inversion of a dynamic piecewise constant curvature (PCC) model using servo-constraints, and the combination of the two different feedforward controllers with a linear quadratic regulator with integral action (LQRi) and gain scheduling are considered. For feedback control, only sensor feedback obtained from a curvature sensor integrated into the soft robot is used. The performance of these approaches is compared in simulations and experiments.Publicationtype: Conference Paper37 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , Turnpikes in Deep Learning: beyond ResNets and neural ODEs?(IEEE, 2026-07); ; ; ; It is well known that deep learning, and in particular the training of ResNets and neural ODEs, can be formalized and analyzed from an optimal control perspective. In this work, we extend the dynamic systems and optimal control perspectives to fully connected neural networks with ReLU activations and no skip connections. By exploiting equivalence relations between ResNets and networks without skip connections, we show that the corresponding training problems exhibit the turnpike property under conditions analogous to those established for ResNets. We illustrate our findings with numerical experiments for the two-spirals dataset and MNIST.Publicationtype: Conference Paper13