BCI Gains Dual Breakthroughs: Faster Learning via Brain Manifolds and Real‑World Long‑Term Use

Human learning of noninvasive brain–computer interfaces via manifold geometry

18 participants performed joystick calibration to identify each subject’s intrinsic neural manifold. Three closed‑loop neurofeedback mappings were tested: intuitive mapping (IM), within‑manifold perturbation (WMP), and out‑of‑manifold perturbation (OMP). Real‑time neural data were re‑embedded onto the personal manifold using a manifold‑regularized auto‑encoder. Participants mastered control of a virtual avatar in under one hour when the mapping followed the natural manifold; no learning occurred with OMP within the same time.

Neural recordings showed reorganization of cortical activity that correlated with individual performance and spread to regions beyond the target area, indicating ripple‑like effects.

Paper link: https://www.nature.com/articles/s41593-026-02311-2

Figure 1: Manifold learning and validation.

Figure 2: Learning performance for different manifold components.

Long‑term independent use of an intracortical brain–computer interface for speech and cursor control

One ALS patient used an intracortical BCI at home for approximately two years. The system consisted of a networked computer mounted on a mobile cart placed in the bedroom or living‑room. A trained caregiver connected the neural recording hardware and power; custom software then automatically initialized the interface, allowing operation without researcher assistance.

Usage statistics: >3,800 hours of independent operation, 183,060 sentences, 1,960,163 words, average 56 words min⁻¹, and >99 % accuracy on a 125,000‑word vocabulary.

Functional outcomes included messaging, email, web browsing, and work communication, demonstrating daily‑life utility.

Paper link: https://www.nature.com/articles/s41591-026-04414-6

Figure 3: Independent use of a multimodal intracortical BCI.

Figure 4: Participant Casey Harrell using the BCI at home for two years.

Emerging focus of BCI research

Competition is shifting from purely model performance to understanding brain structure and real‑world scenarios. Non‑invasive systems aim to respect existing neural geometry and guide training along easier‑to‑generate pathways; invasive systems must integrate performance, stability, and user experience to support sustained independent use.

Future impactful BCIs are expected to combine faster signal acquisition with natural learning curves and reliable long‑term operation, enabling seamless functionality in everyday life.

Code example

来源：ScienceAI
本文
约2000字
，建议阅读
5
分钟
BCI从实验室演示走向现实可持续使用。