BrainCo demonstrated a brain-controlled robot AI platform at WAIC 2026 in Shanghai, reported July 19-20, 2026. A user wearing a lightweight EEG headset thought about grabbing a cap and a robotic arm executed the action. The pipeline runs EEG capture, AI decoding of motor and control intent, then robot commands, with the entire loop under 200 milliseconds.
A decade of BCI research has given us the ability to decode what a person intends to do and translate that into machine action. We believe it will define the next chapter of human-machine collaboration. — Nyx He, partner and SVP, BrainCo
The company brands the framework "neuro-embodied-AI" and describes it as hardware-agnostic, designed for commercially available humanoids, robotic arms and legged robots rather than for a proprietary platform.
The data rig is the quieter announcement
BrainCo also debuted an Embodied AI Data Collection Solution: a dual-arm wheeled platform paired with a high-precision glove, capturing robot execution, human demonstration, simulation — and the operator's EEG. Adding brain signal to a manipulation dataset is a genuinely new axis. Whether it improves policy learning is unproven, but it is a cheap thing to record once the headset is already on.
BrainCo was founded in 2015 and is based in Somerville, Massachusetts. Existing products include the Revo 3 Dexterous Hand, a 21-degree-of-freedom end effector, an Intelligent Bionic Hand five-fingered prosthetic, and an Intelligent Bionic Leg smart prosthetic knee. Demonstration tasks at WAIC included grasping a cup and picking up an apple.
Key Facts
- Demonstrated at WAIC 2026, Shanghai — EEG capture to robot command in under 200 milliseconds end to end
- Framework branded "neuro-embodied-AI"; hardware-agnostic across commercial humanoids, arms and legged robots
- Embodied AI Data Collection Solution: dual-arm wheeled platform plus high-precision glove, recording execution, demonstration, simulation and operator EEG
- BrainCo founded 2015, Somerville, Massachusetts
- Existing products: Revo 3 Dexterous Hand (21 DoF), Intelligent Bionic Hand, Intelligent Bionic Leg
Why it matters
Teleoperation is currently the field's answer to the data problem, and it captures what the operator did but not what they were trying to do. Intent and action diverge constantly — corrections, hesitations, abandoned grasps — and a dataset that cannot distinguish them teaches the model to imitate the noise.
The 200-millisecond figure is the practical constraint. Below roughly a quarter of a second, control feels closed-loop to a human; above it, operators start over-correcting. Getting the whole EEG-to-actuation path inside that window is the difference between a research curiosity and an interface.
Frequently Asked
How fast is the loop?
Under 200 milliseconds from EEG capture to robot command.
Does it need BrainCo hardware?
No — the framework is described as hardware-agnostic across commercial humanoids, arms and legged robots.
Why record EEG during demonstrations?
Teleoperation data captures what an operator did, not what they intended. Brain signal is a new axis on that problem.