The ongoing project started in 2015.
Description
The proposed research aims at developing a novel type of interface between human brains and robots. Brain-robot interfaces (BRIs) are a subclass of so called brain-machine interfaces (BMIs), that is, hardware-software systems that can monitor a person’s brain activity and translate the measured signals into commands carried out by an artificial device (a robot in our case, and any kind of machine in general). In other words, a BRI makes it possible for a human to control a device to some extent directly through intentionally generated brain signals. BRIs/BMIs have a tremendous range of potential applications in various domains such as health care (e.g., support of severely motor-disabled people via wheelchair steering and neuroprotheses), augmented human-computer communication, machine-based manufacturing, and neuromarketing.
In principle various types of BMIs are thinkable, differing in terms of the implemented functional brain imaging method that monitors the neural activation. With respect to BRIs, those functional neuroimaging methods are of particular interest that are portable/mobile and non-invasive. Almost all available BRIs are based on (non-)invasive electroencephalography (EEG), yet with none of these neuroelectric BRIs going beyond basic laboratory demonstrations. Moreover, a considerable subgroup of users is not able to successfully control neuroelectric BMIs. An alternative portable and non-invasive neuroimaging method is functional near-infrared spectroscopy (fNIRS). While no fNIRS-based BRI is available today, it obviously constitutes a promising alternative to EEG-based BRIs. Next to functional magnetic resonance imaging (fMRI), fNIRS belongs to the so-called hemodynamic (vs. neuroelectric) brain imaging. During the last decade, fMRI and more recently also fNIRS have been successfully explored for BMI purposes. However, so far there exists only a single “proof-of-concept” BRI study implementing brain hemodynamics (fMRI), even though hemodynamic brain signals as measured with fNIRS and fMRI have two important advantages as compared to EEG: high single-trial reliability and higher spatial resolution. When combined with intelligent control paradigms , these advantages can enable the development of effective BRI control – even though hemodynamic signals have a relatively low temporal resolution and are detectable only a few seconds after corresponding neuronal events.
Based on these considerations, the proposed research aims at developing a sophisticated brain-robot interface based on fNIRS and exploring its benefits and limitations. This interface will be enriched on the side of the robot with advanced control methods from artificial intelligence (in particular, machine learning-based self-adaptation and autonomous motion/grasp planning) so that a human user gets best possible support in achieving his/her intentions through efficient and direct control of a robot via brain signals. We will first focus on a basic control setting (reduced command set and elementary tasks) and will then iteratively extend this setting by extending the commands and task complexity and successively adding state-of-the-art machine-learning and automated-planning techniques. Most of the technical equipment needed for our research is already available in our research groups, including a mobile fNIRS interface and robots (e.g., turtlebots).
Planned research builds on extensive experience of the applicants in cognitive robotics and neuroscience, including, in particular, intelligent robot control and fMRI/fNIRS-based brain imaging (including BMI).
Support
 |
This project is co-funded by Maastricht University through Luik 3 initiative (Funding for Strategic Innovation).