
多家科研团队攻关“读脑”与“写脑”技术,助力康复医疗与疾病治疗
安徽多家脑机接口团队正推动技术临床应用。爱博智能研发下肢外骨骼机器人实现“读脑”辅助康复,福晴医疗开发无创深脑调控系统实现“写脑”精准干预,已在全国200余家医院应用。
AI-generated summary
脑机接口技术通过采集脑电信号实现人机交互,主要分为“读脑”(意图识别)和“写脑”(神经调控)两大方向。
中新社合肥9月13日电 (记者 张俊)安徽多家脑机接口项目团队近日接受中新社记者采访,探析脑机接口技术如何从实验室走向临床应用。
“要做到患者想动,机器立马就能带着动。”安徽爱博智能科技有限公司产品经理王龙介绍说,其团队研发的脑机接口—下肢外骨骼机器人设备通过采集并解析大脑运动意图,驱动外骨骼执行相应动作,为脑卒中、脊髓损伤等下肢运动障碍人群的康复辅助提供新的技术路径。
王龙解释,所谓“读脑”,就是通过脑电帽采集大脑发出的微弱电信号,从中识别人的运动意图,再转化为机器指令。“脑电信号只有微伏级,还容易被眨眼、肌肉收缩甚至周边电器干扰。要从一团‘噪声’里准确读出人的想法,难度不小。”他表示,团队和中国科学技术大学先进技术研究院成立了脑机接口与康复助残联合实验室,通过自研降噪解码算法破解这一难题。
王龙将技术攻关方向概括为“更快、更准、更稳”。“更快”是让患者“想动”的瞬间,外骨骼或刺激设备能几乎同步响应;“更准”主要是通过自研脑电降噪算法,提升运动意图识别的准确性。“更稳”主要解决在跨时间、跨个体解码时的稳定性问题。
如果说脑机接口—下肢外骨骼机器人是“读脑”的典型应用,无创深脑调控系统则代表了“写脑”方向。
“传统神经调控最深到皮层下2厘米至3厘米,无法到达脑干等深部核团,而帕金森病、阿尔茨海默病的发病机制往往涉及大脑皮层下深部核团。”安徽福晴医疗科技有限公司市场总监张荣介绍,该产品采用时域干涉电刺激技术,可在颅内深部形成精准的低频刺激效果,实现不开颅条件下的精准靶向干预。
张荣介绍说,该系统搭载自研算法,颅脑个性化建模约两小时即可定位患者深部核团,相关经颅电刺激仪已于今年3月获批上市,已应用于全国200余家医院,覆盖精神、康复、疼痛类疾病治疗。该团队正在研发消费级产品,未来有望应用于睡眠改善等场景。
AI outlook — possibilities, not facts
福晴医疗将推出消费级睡眠改善产品
Likely · Within months

Gao Wen, an academician of the Chinese Academy of Engineering, proposed at the 2026 Pujiang Innovation Forum that China's scientific research system should "advance in four directions" by embracing iterative scientific research paradigms, reforming funding mechanisms, strengthening the closed loop of scientific research and manufacturing, and building AI scientific research infrastructure to gain advantages in the new round of scientific research competition.

Chen Tsz-kin, a 17-year-old in Hong Kong with Duchenne muscular dystrophy, has unveiled an AI version of himself. The digital twin is designed to share his life story and message of perseverance after he passes, defying his prognosis of an 18-year life expectancy.
During the 2026 Pujiang Innovation Forum, Shanghai’s “Pearl Constellation Project” was officially launched. The plan has the long-term goal of building a "gigawatt-level space-based computing constellation" and aims to build a space-based intelligent computing infrastructure with global coverage and space-ground coordination by 2030, and promote the extension of intelligent computing capabilities to low-Earth orbit.

The Japanese government adopted the "Generative AI Intellectual Property Protection and Transparency Principles" on August 25, requiring AI developers to disclose the outline and collection methods of training data. This move aims to solve the problem of information asymmetry, promote AI copyright governance from "permission or prohibition" to "transparency obligations", and seek a balance between industrial development and rights protection.
The 2026 AI-empowered manufacturing green development themed event was held in Ningbo. The event released the "Ningbo Green and Low-Carbon Manufacturing One-stop Service Platform" and the first batch of AI industrial application products, aiming to optimize the production process through artificial intelligence technology, reduce energy consumption and pollution emissions in the manufacturing industry, and promote the green transformation of the industry.

Chinese researchers developed a method to increase the durability of wurtzite ferroelectric memory chips, achieving over 10 billion writing cycles—roughly 100 times previous endurance levels—to advance next-generation computing hardware.