1) The brain signal (most often EEG) is recorded continuously, without requiring deliberate actions from the user. 2) Neurophysiological features correlated with mental states are extracted (e.g. band power, event-related potentials, ErrP). 3) A machine-learning model classifies the user's mental state in real time (e.g. high/low cognitive load, attention, error detection). 4) The result is passed to the system, which adapts its behavior (a neuroadaptive loop), e.g. simplifying the interface under high load or correcting behavior after detecting error perception.
Classic (active/reactive) BCIs require deliberate effort and commands, which is tiring and has low throughput. Passive BCI lets systems understand the user's state without extra burden, enabling natural, adaptive interaction.
Continuous recording of the brain signal (electrodes + amplifier), without requiring deliberate user actions.
Official
A machine-learning model recognizing mental states (load, attention, ErrP) in real time from signal features.
Official
A mechanism that uses the detected mental state to adapt the behavior of the system/interface.
Official
The EEG signal is weak and noisy; artifacts (motion, blinks, EMG) reduce reliability.
EEG patterns differ across people and sessions, often requiring per-user calibration.
Thorsten O. Zander and Christian Kothe introduce the notion of passive BCI and the active/reactive/passive taxonomy.
Zander Labs is founded, developing neuroadaptive technology based on passive BCI (the SAMANAI platform, the Zypher EEG suite).
Time complexity: O(kanaly × okno × cechy) na klasyfikacje. Space complexity: O(model klasyfikatora + bufor sygnalu).
Low signal-to-noise ratio (SNR), artifacts (motion, blinks) and inter-individual variability make reliable real-time classification difficult.
Which state is detected (load, attention, engagement, ErrP).
Source of the neurophysiological signal (most often EEG).
Extent of per-user calibration vs user-independent models.
Continuous, passive signal processing; the output depends on the user's mental state.
Continuous monitoring, no routing.
Channels and features can be processed in parallel; the control loop is sequential.
Mental-state classification from EEG is lightweight and can run on CPU / embedded devices (local processing).
