From Disable to Able: How Aible's Non-Invasive BCI Restores Movement with AI

Company Vision and Name

Aible's English name combines "AI" and "Able," reflecting its mission: using AI to help people with disabilities regain capabilities — removing the "dis" from "disable." The company first applies this vision to rehabilitation, where a patient's movement intent from the brain may not reach the limbs due to neurological damage.

Technical Challenge: Non-Invasive BCI for Rehabilitation

EEG signals collected from the scalp are extremely weak and contaminated by eye movements, muscle activity, and environmental noise. Even after algorithmic decoding, the intent must be translated into safe, executable actions by a robotic exoskeleton. The core question: can algorithms reliably interpret EEG in real hospital environments?

Invasive vs. Non-Invasive BCI: No Superior Route

General Manager Su Ying states that neither invasive nor non-invasive BCI is inherently better. Invasive electrodes capture clearer signals but require surgery; non-invasive EEG avoids surgery and lowers adoption barriers but suffers from low signal-to-noise ratio and high susceptibility to artifacts. Both routes face unresolved challenges. Su emphasizes that BCI remains a "future industry" with no converged technical standard. Two drivers accelerate progress: material advances improving invasive electrode quality and throughput, and AI providing new tools for complex signal processing and cross-subject generalization.

Why Non-Invasive? Market and iFlytek Heritage

Aible chose the non-invasive path for two reasons: a broader addressable market and the signal-processing expertise of its parent company, iFlytek. As a pioneer in deep learning for speech denoising and recognition, iFlytek offers a transferable engineering path. Speech and EEG are both continuous time-series signals requiring denoising and decoding, but non-invasive EEG is weaker, more noise-prone, and exhibits greater inter-subject and intra-subject variability. To tackle this, Aible co-founded a joint lab with Professor Chen Xun's team at the USTC Institute of Advanced Technology: the university contributes research foundations and prototypes, while Aible handles end-to-end software/hardware engineering and clinical deployment.

Rehabilitation as First Landing Scenario

EEG enables both state assessment and device control, forming a clear "central-peripheral-central" closed loop in rehabilitation: the patient generates movement intent → EEG device captures and decodes → rehabilitation robot executes movement → real proprioceptive feedback plus visual/auditory cues return to the nervous system → positive neuroplasticity is induced. Aible's products currently target stroke and spinal cord injury patients. Early robotic training is often passive (pre-set gait patterns); BCI adds an active pathway by detecting movement intent before the body moves, allowing the exoskeleton to assist in sync with the patient's own neural command.

Product Architecture: EEG, Decoding, Exoskeleton

Aible's lower-limb gait training device has obtained a medical device registration certificate and is already in hospital use. It provides the engineering foundation and the actuation endpoint. The critical link is the middleware algorithm that continuously processes EEG and controls the robot. A typical session: therapist enters patient data → exoskeleton fitted for size and assistance level → patient performs guided motor imagery (e.g., walking) while wearing an EEG cap → signals are acquired, processed on-edge, decoded into movement intent → commands sent to exoskeleton → patient walks with assistance. This completes a full neuroplasticity-reinforcing loop.

Three Engineering Goals: Faster, More Accurate, More Stable

Faster: End-to-End Latency Under 80 ms

Su Ying contrasts a common piecemeal architecture — EEG device → host PC → robot vendor software → actuator — with Aible's integrated edge approach: EEG data processed directly on an edge AI chip, invoking gait algorithms and robot control locally. The target end-to-end latency is <80 ms to avoid intention-action dissociation that would break the neurofeedback loop. This relies on a streaming lightweight real-time decoding architecture and domestic high-performance edge computing to eliminate cloud transmission delay.

More Accurate: Noise Removal and Decoding Precision

Accuracy refers to movement intent recognition; errors (e.g., classifying "extend leg" as "lift leg" or "no movement") not only waste training but can induce harmful erroneous neurofeedback. Non-invasive EEG's low SNR demands rigorous artifact separation (eye blinks, EMG, movement artifacts) before decoding. Professor Chen's lab has developed a series of joint blind source separation and deep learning denoising algorithms. Aible's engineering team simultaneously optimizes these models for on-edge compute and continuous operation. Su stresses that "more accurate" starts with cleaner inputs to reduce downstream decoding drift.

More Stable: Cross-Subject Generalization and Adaptive Alignment

Stability addresses two variability sources: (1) Inter-subject differences — EEG features vary across individuals. Aible explores a shared-representation approach: extract stable common features from a cohort of stroke patients to build a universal decoder, then perform few-shot personal calibration for each new patient, avoiding per-subject training from scratch. (2) Intra-subject day-to-day changes — electrode placement, fatigue, and mental state shift. The team researches adaptive alignment to adjust model parameters to the day's signal state. "Stable" also includes algorithmic safety: a reject-option strategy withholds unreliable decodings from becoming commands, and the exoskeleton only executes within pre-defined safe gait envelopes.

Clinical Deployment at Anhui Provincial Second People's Hospital

In February 2024, the hospital (a Grade-A tertiary hospital) opened a dedicated BCI clinic and ward, accumulating 250+ complete clinical cases. The Anhui Provincial AI Scenario Project "Non-Invasive BCI-based 'AI+Rehabilitation' Innovative Scenario Development and Demonstration" was launched in August 2026 with Aible as the winning bidder. The hospital built a BCI ward; Aible stationed engineers on-site. On-site deployment provides invaluable real-world feedback: patient physiology fluctuates, therapists have established workflows, and equipment faces ward environmental stresses and continuous-use pressure. This data also fuels algorithm development — cross-subject models need real patient EEG; adaptive alignment and reject strategies require multi-day, multi-interference validation. The project scope extends beyond motor rehabilitation to attention, emotion, and cognitive disorders, using multimodal data (EEG, eye-tracking, scales, behavior) for each population with separate data collection and clinical validation.

Founder's Philosophy: Clinical Validation Over Hype

Su Ying, a public health graduate and former CDC officer, identifies as a "medical person." Her guiding principle: "Entia non sunt multiplicanda praeter necessitatem" (Do not multiply entities beyond necessity). Aible focuses on rehabilitation first, perfecting "faster, more accurate, more stable" BCI rehab products — better denoising, better decoding, better generalization — before expanding to elderly/pediatric care, education, industry, and embodied AI. She argues startups should invest limited resources in work that yields deployable products and closes the minimum business loop. Chasing "first" or "number one" labels is meaningless if the product fails in real clinical use; only actual usage data and cases spin the data-algorithm flywheel. This ethos extends to user dignity: at exhibitions, staff must squat to eye-level with wheelchair users — a small gesture reflecting whether technology truly serves people.

Conclusion: Technology Must Serve Human Dignity

Su believes everyone may face capability limitations from disease, aging, or temporary conditions. "Able" does not mean restoring a uniform "normal" but helping a person regain a specific lost function. BCI offers a new tool; its real-world stability on patients must be proven in hospitals. The invasive vs. non-invasive future remains open; Aible picked one lane and built its starting point in rehabilitation. The ultimate measure is not technical jargon but whether people can use it safely and effectively, and whether impaired functions actually recover and improve. From Disable to Able, there are no shortcuts — the product must pass the test of real human lives.