Visit by the Elite Network of Bavaria

On July 9th 2019 the Neuroengineering Master Program at TUM was visited by representatives from the Elite Network of Bavaria (ENB) as well as several renowned external professors in neuroscience and -engineering. Since we are supported by both, the master program and the ENB, we were happy to present our team and our current research with a demo booth at the event.

Our exchange student Jasmine from the University of British Columbia presented her work on real-time temperature feedback. This is challenging because off-the-shelf temperature sensors take some time (~1min) until they register a stable temperature signal. This is of course absolutely insufficient for the use in prostheses. Imagine putting a prosthesis hand on a hot stove and realizing it a minute later... Our idea is to use machine learning to predict future temperature measurements as early as possible. The predicted future temperature value is then sent to a wearable cooling and heating patch. For safety purposes we re-map the measured/predicted temperature to a non-harmful interval. It is not our intention to directly feedback a predicted temperature of 100°C to the user and thus burn her or him, but instead just enough so that they perceive the stimulus as harmful and retract their hand quickly.

Valerie, our TUM PREP summer intern from MIT, showed a demo of using one-shot-learning for camera-based grasp planning for prostheses. The project which is supervised by Zied is inspired by previous research from Weiner et al. at Tamim Asfour's lab at KIT. The basic idea is to recognize the object in front of the prosthesis and use a neural network to predict the most appropriate grasp type (e.g. a pinch grasp or a power grip). The prosthesis user then only has to control whether she or he wants to grasp the object or not (binary control signal), instead of having to control every finger joint individually (high-dimensional control signal). We believe that one-shot-learning is a very promising approach for quickly adding new, previously unseen, object types and associated grasp types to the pre-trained neural network model during operation of the prosthesis.

We also showed a 3D printed open source prosthesis and the very first prototype of an arm prosthesis developed from scratch by multiple team members supervised by Johannes and Alex.

In a presentation to the guests, Nicolas presented the paradigm of human-centered engineering and how it can be used for neuroengineering and robotics projects such as those that the Cybertum Student Team is working on.

Nicolas Berberich

Photos: Nicolas Berberich