High-performance brain-to-text communication via handwriting

The 60-second view

What problem?

Point-and-click typing capped BCI communication well below everyday typing speeds.

What did they do?

Decoded attempted handwriting movements from motor cortex into text in real time with a recurrent neural network.

What changed?

Roughly doubled the best previous BCI typing rate and showed that complex, fast movements can be easier to decode than simple ones.

Why does it matter?

It changed the field's intuition about what to decode: rich, time-varying movements carry more information than cursor trajectories.

Method

Participants

One person with hand paralysis from spinal cord injury

Interface

Two intracortical microelectrode arrays in motor cortex

Signal

Intracortical spiking activity

Task

Attempted handwriting of letters and sentences

Decoder

Recurrent neural network, with optional autocorrect

Results

90 characters/min

Typing rate

94.1%

Raw accuracy, online

>99%

Accuracy with autocorrect, offline

Strength of evidencePeer reviewed, one participant; dataset publicly released.

Figures are quoted from the published abstract and were checked on 3 Oct 2026.

Limitations

• One participant.
• Requires periodic decoder recalibration.

What this could enable

Decoding of rapid, dexterous movement years after paralysis, and the sequence-decoding methods later used for speech.