ML for brain-machine interfaces
We investigate various methods to control robotic assistive devices via brain interfaces. We use surface EMG for controlling the grasp force of the hand as well as the position of the arm in a user-intuitive manner. To this end, an adaptive system learns the correspondence between the human hand position and orientation and the muscular activity measured at the skin surface. Thereby we can control to move the arm and grasp an object in teleoperation / telemanipulation. Such control schemes are also applicable in rehabilitation and orthoses environments.
EMG and EEG
We use various machine-learning methodologies for processing peripheral neural data. In particular, we focus on recurrent neural networks and variational autoencoders (VAE) to process and preprocess these data.
We are furthermore investigating the control of such robots through human cortical implants. In this control scheme, neural signals recorded in the human motor cortex are decoded in continuous motion commands, that are executed by the robot. This combination of state-of-the-art robotics and advanced neuro-prosthesis allow a person with severe physical disabilities to physically interact with their environment again. Our results have been published in Nature in 2012. Further information can be found here.
Other interfaces, including invasive communication with the human peripheral nervous system as well as surface EEG control are within the realm of our research spectrum.