Context-aware self-calibration

A method for autonomous control of assistive devices comprising:

• measuring for a plurality of time segments, at least one feature of at least one neuronal signal for each of a plurality of channels, each channel associated with at least one neuron of a user, each neuronal signal created by the at least one neuron;
• decoding the at least one neuronal signal to move an object along an actual decoded direction determined by an output of the decoder configured to correlate for each time segment, wherein the user imagines an intended direction, the at least one feature to the intended direction;
• calibrating the decoder by minimizing an error between the actual decoded direction, and the intended direction inferred subsequent to the respective time segment;
• normalizing the at least one feature during a rest period, wherein the user refrains from imagining the intended direction, by estimating a baseline value of each feature with a recursively-defined exponential estimate, the baseline value of each feature being subtracted from each respective feature for each time segment before being decoded by the decoder;
• and
• correcting a velocity bias, wherein the user imagines the intended direction, by determining an exponentially weighted running mean of a decoded velocity for each respective time segment exceeding a speed threshold, and subtracting the exponentially weighted running mean from the decoded velocity.

A system for autonomous control of assistive devices comprising:

• a user imagining an intended direction for each of a plurality of time segments by generating at least one biosignal from a part of the user;
• a biosignal sensor configured to measure for the plurality of time segments, at least one feature of the at least one biosignal for at least one channel;
• a device electrically decoupled from the part of the user;
• a decoder configured to correlate for each time segment, the at least one feature to the intended direction, wherein an object is moved along an actual decoded direction determined by an output of the decoder, and the decoder is calibrated by minimizing an error between the actual decoded direction, and the intended direction inferred subsequent to the respective time segment;
• and
• wherein each feature is normalized during a rest period, including refraining by the user from imagining the intended direction for at least one time segment, and a baseline value of each feature is estimated with a recursively-defined exponential estimate, the baseline value of each feature being subtracted from each respective feature for each time segment before being decoded by the decoder.

A method for context-aware self-calibration comprising:

• measuring for a plurality of time segments, at least one feature of at least one biosignal, each biosignal created in response to an intended action for each time segment;
• performing an actual action determined by an output of a decoder configured to correlate for each time segment the at least one feature to the intended action;
• calibrating the decoder to minimize, for each time segment, an error between the actual action, and the intended action inferred subsequent to the respective time segment;
• and
• normalizing each feature during a rest period, including refraining by a user from imagining the intended action for at least one time segment, and estimating a baseline value of each feature with a recursively-defined exponential estimate, the baseline value of each feature being subtracted from each respective feature for each time segment before being decoded by the decoder, and wherein the user generates each biosignal.