In this project, the landing position of a sandbag is to be predicted with the help of AI. To do this, a data set is recorded with the ElaasticNode attached to the lower arm. This measures the acceleration with a sensor. The landing position is evaluated via a camera. A data set can be built from this. A neural network is then to be trained on this data set. This neural network will then be transferred to the ElasticNode using the ElasticAi.Creator and evaluated.

This results in the following points that can be worked on:

• Preparation and execution of the data set recording
• Finding the best model for landing position prediction
• Designing an extension board for the ElasticNode with a different sensor
• Local training on the ElasticNode to adapt to the respective user

Organisation:

Attendance at the kick-off meeting is mandatory for participation in this project.

As part of this practical project, students are to optimise the methods for neurosignal processing of extracellular action potentials, which are recorded using microelectrode arrays.

A Python framework already exists for this purpose, which is to be expanded with additional functions for AI-based methods, additional functions for synthetic data generation and for neuronal data analysis (incl. representation). For this purpose, the classification tasks are to be validated using deep learning techniques and with neuromorphic networks via spiking neural networks. In addition, there is the possibility to further optimise the internal hardware setup for the playback of neurosignals from digital source to analogue signal.

Possible subjects:

• Data set creation with MEArec
• Data analysis with MEAnalyzer
• Preparation of data sets for autoencoder training (Dense NN, CNN, Denoising, ...)
• Use of the elasticAI.Creator to generate neural networks
• Use of neuromorphic networks
• Preparing the Neurosignal Player (C code for playing the signals)

Organisation:

Attendance at the kick-off meeting is mandatory for participation in this project.