Whole-Body Control for Legs with Closed-Loop Linkages
Most of the quadrupedal robots we see today have legs that, essentially, consist of two links (see picture below). There have been a few exceptions (e.g. the MIT Cheetah 2) that used more complex linkage structures. However, in these cases, the linkage structures have typically been chosen such that 1) those links that couple degrees of freedom are very lightweight and 2) the resulting closed-loop chains exhibit the kinematics of a parallelogram – one of the few closed-loop structures for which analytical solutions are known. Therefore, in these cases, one could get away without really accounting for the closed-loop linkages in the control algorithms. Accordingly, as far as we know, there currently does not exist any implementation of a hierarchical/prioritized whole-body controller (see the paper in the application paragraph below) that could handle general closed-loop linkage structures (for which analytical solutions typically do not exist). Yet, the ability to control legs with arbitrary closed-loop linkages would allow for significantly more flexibility in the mechanical design of robotic legs. The introduction of additional coupled links could, for instance, be used to increase energy efficiency, reduce the envelop (avoiding collisions is the reason why robots such as Spot walk down stairs backwards), or improve the torque-to-force ratio for certain configurations. Advantages might also arise from different designs of the front and the hind legs, for instance when jumping. We would thus like to extend hierarchical whole-body control to efficiently handle arbitrary closed-loop linkages and use it to subsequently investigate novel, improved leg designs.
Mechanical Design of a Versatile Arm for Legged Robots
Mechanical Design of a Versatile Arm for Legged Robots
Augmented Virtuality
While immersed in a virtual reality, we are usually oblivious to the real world. Augmented virtuality is a concept where the virtual world is enriched with real physical objects, or virtual representations of them. The goal of this project is to design an augmented virtuality system using a combinat
Deep learning Hololens project in collaboration with Microsoft
Microsoft has proposed two projects related to object pose estimation and interaction with robots. More details in the attached document.
HoloAssembly: Learn an assembly task using HoloLens
One of the most exciting uses of mixed-reality head-mounted devices like the Microsoft HoloLensis to assist a user to learn / perform tasks.The HoloLens sees the world through a depth and a RGB camera, and can map the position of theuser in the real world through the built-in head tracking. It can also provide compelling experienceswhere virtual 3D objects (holograms) are mixed with real ones in the real world. However, the devicedoes not have capabilities yet to reason about what the user is doing, or how they should interact withthe objects around. The goal of this project is to start exploring this area, by building a HoloLensapp to assist the user in a pre-determined assembly task – like assemblying a piece of Ikea furniture.The student will get access to a HoloLens 2 device, and will therefore be able to leverage recentcapabilities like fully articulated hand tracking and eye gaze tracking.
Simulation and parameter estimation for soft robotics
We aim to build fast simulation models of soft deformable materials in order to facilitate soft robot design, or robotic manipulation of elastic objects.

Powered by  SiROP - the academic career network