Non-invasive brain-computer interfaces

Prof. Miriam Zacksenhouse | Mechanical Engineering


Information and Computer Science | Medical Devices

The Technology

Non-invasive Brain Computer Interfaces (BCIs) measure electroencephalography (EEG) potentials from the scalp and interpret them, through signal processing and machine learning, to make decisions and operate computers or other external devices. Error-related potentials are produced by the brain when a person makes a mistake or observes erroneous actions, such as unexpected movements of a robot/exoskeleton, visual and haptic disturbances or BCI mistakes. They offer a valuable insight into investigating how the brain processes errors. Error-related potentials can be monitored to categorize actions as correct or erroneous. Detection of erroneous movement can be used to improve BCIs, detect task-relevant erroneous movements, identify instances of motor errors, and enhance motor training, rehabilitation, and other applications. Common approaches involve binary classification of error-related potentials (error/no error or error A/error B).

A novel approach has been developed for multi-class classification of error-related potentials to distinguish between no error/error A/error B, and so on. Multi-class classification of error-related potentials is critical for:

  1. Developing self-corrected BCIs, where the detected type of error is used to correct the corresponding mistake made by the BCI. This is critical since the main challenge in using BCIs is their low accuracy.
  2.  Automatic annotations of erroneous and dangerous situations by detecting error-related potentials in observers. Observers may view the situations in real time or in videos. Annotation is a critical and time-consuming stage in training AI systems to automatically detect predefined events. Utilizing error potentials for annotation of erroneous/dangerous situations is critical for real time observations and can facilitate video annotation. The most important advantages of using EEG-based annotation, compared to manual annotation, is the synchronization between the event and its annotation and the reduced mental load on the observer.


  • Accurate and well-synchronized annotation
  • Less strenuous process

Applications and Opportunities

  • Annotation of erroneous/dangerous situations for training AI and ML systems, including training of autonomous driving, and virtual trainers.

  • Self-correcting BCIs for commanding different applications including virtual reality interfaces and augmented reality headsets.

arrow Business Development Contacts
Shikma Litmanovitz
Director of Business Development, Physical Science