Non-invasive brain-computer interface startup secures angel funding as it builds ‘intent layer’ for machines to understand human thought before action
By Zan Zhu
A person dons a minimalist neural interface headset — no conductive gel required. As their attention shifts, a fleet of drones seamlessly changes formation. Next, the same neural control interface switches to a robotic arm to grasp an object. Moments later, a robotic dog moves and performs a retrieval task.
This is not a scene from science fiction. It is a controlled demonstration recently conducted by Wabo Technology, founded in 2025 by Zhang Xingzhi who was previously chief architect for smart manufacturing at Tencent Holdings.
The non-invasive brain-computer interface (BCI) company recently announced the close of its latest funding round. The angel round saw participation from Singapore-based Temasek-backed Vertex Ventures, Hongtai Capital, Future Industrial Fund, and Huifengda Capital.
What convinced investors to dive in was not its lighter electroencephalography (EEG) hardware or more impressive demonstrations of BCIs, but a vision — once the stuff of science fiction but now broken down into engineering realities by Wabo — to build a neural intent model. This model enables machines not only to see what a person has done but also, before an action occurs, to understand what the person intends to do, whether they have confirmed the action, when they are correcting an error, and when they must stop.
If large language models seek to enable machines to understand human language, Wabo’s neural intent model aims to supply the missing “human intent layer” required for artificial intelligence to operate effectively in the physical world.
Moving BCI out of the lab
BCI research has traditionally been defined by metrics in academic papers, channel counts, and laboratory-grade precision. Heads covered in wires and conductive gel, lengthy calibration processes, and the need for professional assistance—these elements formed the standard image of a BCI demonstration.
Wabo claims to have made breakthroughs across multiple application scenarios, leveraging full-stack system capabilities — spanning neural acquisition interfaces, signal processing, neural intent recognition models, and data governance. In a recent interview with state television, Wabo and fellow Chengdu-based BCI enterprise Gestalt Technology (Gestala) its technology by using a lightweight, non-invasive neural interface to control multiple types of machines, including drone formations, robotic arms, and quadruped robots in real-world settings.
The startup says its Neural Intent Data Platform (NIDP) has completed compatibility testing with Huawei’s Ascend AI processors and obtained Ascend technical certification. It has established a complete operational pipeline spanning neural signal acquisition, data processing, model training, edge inference, and physical control.
Wabo also said it has spent years optimizing components across the hardware stack, including electrodes, analog front-end chips, data acquisition boards, filtering and interference suppression, device design, algorithms and control interfaces. It has worked with upstream semiconductor and supply-chain partners to localize key parts of its signal acquisition chain.
What investors are betting on
AI can already recognize images, understand language, and mimic human actions from video. Yet in real-world physical tasks, machines remain deprived of a critical type of essential information: why a person performs a given action.
A camera can see a hand reaching toward a cup, yet struggles to determine whether the person intends to drink, hand it to someone else, or move it away upon detecting danger. A machine can learn grasping trajectories but cannot tell when a person confirms a result, senses something wrong, or wants it to stop immediately.
This is the problem the neural intent model seeks to solve.
Wabo’s neural intent model is not about reading a person’s every thought independently of the task at hand, nor is it an AI wrapper around a few EEG classification commands. Instead, it combines EEG signals, electromyography (EMG), behavioral data, visual information, device status and execution results within defined task and environmental conditions to infer a user’s goals, action tendencies, confirmation, correction, stopping or takeover intentions.
The model’s output is not merely “left” or “right,” but a set of machine-interpretable intent vectors encompassing objective, action, state, confidence, uncertainty, and whether execution should be rejected. What this model must first learn is not even what to do — but when to do nothing.
The data flywheel
Wabo says its neural intent model enables neural technology, for the first time, to enter broad, real-world human scenarios at scale, allowing machines to move beyond imitating human behavior toward understanding tasks and collaborating with people.
As its consumer devices, including products branded Lingxi and Zhimengzhe, reach more users and scenarios, Wabo can accumulate neural data — with user consent —across individuals, time periods, and labeled with task conditions and execution outcomes.
To this end, Wabo has built its own platform for collecting, cleaning, governing and training neural datasets that integrates neural signals with task events, timestamps, device status, signal quality, model outputs and execution results. Following quality control, de-identification and governance procedures, authorized data is used to train and validate its models before updated software is deployed back to devices.
Wabo said the process creates a long-term feedback loop in which better products encourage wider adoption, generating richer datasets that further improve model performance.
From Chengdu to Shanghai
Alongside the funding round, Wabo has completed its landing arrangements with Shanghai’s Minhang District government and the “Brain-Intelligence World” initiative. The company will officially establish operations as a Fudan University-incubated BCI innovation center project and plans a dual-headquarters structure in Chengdu and Shanghai. Shanghai will focus on advanced R&D, neural intent models, closed-loop neural modulation, and industry-academia-research-medicine collaboration. Chengdu will focus on hardware and product development, supply chain, engineering delivery, and large-scale scenario deployment.
Wabo argues that as the human intent model is trained and enhanced, the brain-intelligence era will arrive — pointing toward a more natural human-machine symbiosis: humans need not control every joint of a machine individually but provide goals, preferences, confirmations, and corrections, while an “other body” performs perception, planning, and execution with AI assistance — keeping ultimate control in human hands.
Source:
ChinaVenture
