Register now After registration you will be able to apply for this opportunity online.
Controlling a Magnetically Actuated Inverted Pendulum
Balancing a 3D inverted pendulum using remote magnetic actuation
Keywords: Inverted Pendulum, Machine Learning, Feedback Control, Dynamical Systems
Remote magnetic manipulation, or magnetic actuation, uses magnetic fields to wirelessly control the positioning and orientation of magnetic objects. The electro-Magnetic Navigation System (eMNS) we developed can precisely manipulate a magnetic object's spatial orientation by generating torques and forces through alterations and modulations of the magnetic field and its gradients, respectively.
Our eMNS's dynamic capabilities have been demonstrated by stabilizing a 3D inverted pendulum on a magnetically driven arm. This achievement, which can be viewed in a video (https://youtu.be/fNWS-9-lD84), showcases our system's ability to reject disturbances and follow trajectories using an advanced iterative learning controller. For this work, we have used the magnetic field vector as a control input, hence not harnessing the full potential of all eight coils in the OctoMag system. However, there is great potential in leveraging the over-actuation in the eMNS and include magnetic field gradients into the modeling and control.
Remote magnetic manipulation, or magnetic actuation, uses magnetic fields to wirelessly control the positioning and orientation of magnetic objects. The electro-Magnetic Navigation System (eMNS) we developed can precisely manipulate a magnetic object's spatial orientation by generating torques and forces through alterations and modulations of the magnetic field and its gradients, respectively.
Our eMNS's dynamic capabilities have been demonstrated by stabilizing a 3D inverted pendulum on a magnetically driven arm. This achievement, which can be viewed in a video (https://youtu.be/fNWS-9-lD84), showcases our system's ability to reject disturbances and follow trajectories using an advanced iterative learning controller. For this work, we have used the magnetic field vector as a control input, hence not harnessing the full potential of all eight coils in the OctoMag system. However, there is great potential in leveraging the over-actuation in the eMNS and include magnetic field gradients into the modeling and control.
The first phase of the project will enhance the existing dynamic model and control structure to integrate magnetic field gradients. This will enable a more nuanced control, allowing for refined manipulation of the magnetic object in space.
The second part of the project can be adjusted to the student’s interests. Possibilities are:
Designing a swing-up maneuver that autonomously positions the pendulum in its equilibrium position, demanding a deep dive into non-linear dynamics and optimal control theory; or
Developing learning-control algorithms that employ Neural Networks to estimate the complex dynamics between the magnetic field and the pendulum system, potentially expanding the operational radius.
The first phase of the project will enhance the existing dynamic model and control structure to integrate magnetic field gradients. This will enable a more nuanced control, allowing for refined manipulation of the magnetic object in space.
The second part of the project can be adjusted to the student’s interests. Possibilities are:
Designing a swing-up maneuver that autonomously positions the pendulum in its equilibrium position, demanding a deep dive into non-linear dynamics and optimal control theory; or Developing learning-control algorithms that employ Neural Networks to estimate the complex dynamics between the magnetic field and the pendulum system, potentially expanding the operational radius.
Please email your CV and transcripts to zjasan@ethz.ch
Please email your CV and transcripts to zjasan@ethz.ch