Skip main navigation
We use cookies to give you a better experience, if that’s ok you can close this message and carry on browsing. For more info read our cookies policy.
We use cookies to give you a better experience. Carry on browsing if you're happy with this, or read our cookies policy for more information.

The challenges of controlling robots using our thoughts

In recent years, BCI research has lead to a large variety of proof-of-concept BCI systems. However, despite this impressive expansion, none of the BCI systems described in the literature are sufficiently mature for daily use out of the laboratory.

To make BCI an intuitive interface for use in either control, communication or rehabilitation, improvements are required in several areas, such as usability, signal acquisition techniques, hardware development, machine learning and signal processing, and system integration (Vaadia, 2009).

Generally, in BCI, brain activities are recorded by EEG due to its high temporal resolution, non-invasiveness, relatively low cost, and portability. Despite these advantages, dealing with EEG signals is a challenging task.

Noise in the nervous system

The EEG signals are very noisy. The information of interest is normally mixed up with several irrelevant but concurrent neural activities called central nervous system (CNS) noise. CNS noise may interfere with the information of interest due to overlapped characteristics. For example, BCI systems that work based on imagination of body movement involve the brain patterns in the μ-rhythm (i.e. 8-13Hz) (Pfurtscheller and Lopes da Silva, 1999). However, the α-rhythm, which represents the visual and mental efforts, has the same frequency range as the μ-rhythm. Although the α-rhythm originates from the occipital lobe of the brain, due to the volume conduction, it interferes with the μ-rhythm that originates from the motor cortex (Nunez et al, 1997).

Besides, the α-rhythm is quite prominent, whereas the μ-rhythm is weaker and can only be observed after appropriate signal processing.


Another problem related to classification of EEG signals is non-stationarity. Variation in the EEG properties, called non-stationarity, can result in deteriorated BCI performances as most of the machine learning algorithms implicitly assume stationary data.

This means if the model is fitted to the subject’s training data, it may not be optimal for a new session (recorded on another day) or even new trials in the same session (Kawanabe et al, 2010). The plausible reasons for the changes in the dynamic of the EEG signals are as follows:

  • The physical properties of the EEG electrodes change over time. For example, the impedance of the electrodes can change due to drying up of the conductive gel, or the position of the electrodes can change when the EEG cap is reused in a new session.
  • Neurophysiological conditions (e.g. fatigue) can be variable over time.
  • Psychological parameters, such as motivation, attention and task involvement also display a large variability.
  • Artifacts caused by body movements or muscular activities such as swallowing or blinking change the signal properties.

These drawbacks can clearly impede the continuous use of the BCI system. BCIs in real-life situations must be accurate and robust in their application. A patient needs a system that functions accurately all the time and is able to adapt to new circumstances. Therefore, it is necessary to investigate these issues and find solutions for improving the accuracy and robustness of the BCI systems.

The solutions include developing advanced signal processing and machine learning algorithms as well as training users in generating stable brain patterns over time using feedback training approaches.


Kawanabe M, Pascual J and Vidaurre C (2010). Investigation of Non-stationarity in Brain Activity via Robust Principal Component Analysis. Front. Comput. Neurosci. 4(46), pp.1-5.

Nunez P. L., Srinivasan R., Westdorp A. F., Wijesinghe R. S., Tucker D. M., Silberstein R. B., et al. (1997). EG coherency I: statistics, reference electrode, volume conduction, laplacians, cortical imaging, and interpretation at multiple scales. Electroencephalogr. Clin. Neurophysiol. 103, pp.499–515.

Pfurtscheller, G. and Lopes da Silva, F. (1999). Event-related EEG/MEG synchronization and desynchronization: basic principles. Clinical Neurophysiology, 110(11), pp.1842-1857.

Vaadia, E. (2009). Grand challenges of brain computer interfaces in the years to come. Front. Neurosci., 3(2), pp.151-154.

Share this article:

This article is from the free online course:

Building a Future with Robots

The University of Sheffield

Get a taste of this course

Find out what this course is like by previewing some of the course steps before you join:

Contact FutureLearn for Support