AI News AI资讯 4h ago Updated 1h ago 更新于 1小时前 47

Shenzhen Embodied AI Company Secures Billion-Yuan Financing from Inovance and China Telecom; ‘Visual-Tactile’ Sensor Shipments Lead the Industry | Hardcore First 深圳具身公司获得汇川、中国电信亿元融资,“视触觉”传感器出货量行业第一|硬氪首发

A hundred-million-dollar investment has been directed toward a little-known robotics company, betting on "tactile sense"—the most fundamental and yet instinctively captivating of human abilities. Daimon Robotics has secured the funding, but what’s even more intriguing is that it has touched a nerve in the collective anxiety of the embodied intelligence industry: behind all the flashy demos, robots must ultimately make real contact with the sticky, hard, and slippery physical world. The current i 亿元融资砸向一家名不见经传的机器人公司,赌的是“手感”这个最基础、也最性感的人类本能。戴盟机器人拿到钱了,但更值得琢磨的是,它戳中了具身智能行业集体焦虑的那个点:所有炫酷的demo背后,机器人终究要和这个黏糊糊、硬邦邦、滑溜溜的物理世界发生真实接触。目前的行业逻辑,正从“跑得稳”、“看得见”的炫技阶段,被迫切换到“摸得准”、“握得住”的求生阶段。这场切换,不是升级,而是补课。

70
Hot 热度
68
Quality 质量
62
Impact 影响力

Analysis 深度分析

A robot built solely on a vision-based approach is like a giant infant with severe visual impairment and tactile dysfunction. It can see an egg but doesn’t know it’s fragile; it can recognize a sponge but doesn’t understand the mechanics of compression and rebound. CEO Duan Jiangya’s metaphor is spot-on: vision tells you where an object is, but touch tells you what happens when you make contact. This lack of "physical common sense" is the biggest chasm preventing laboratory robots from entering everyday homes. Daimon’s bet is essentially reinventing a "baby-like" cognitive approach for robots—building a world model through touch. The idea is both archaic and radical.

Yet the barriers to building a tactile world are far higher than developing a suite of vision algorithms. This isn’t just about adding a sensor. Daimon’s claimed "data flywheel" and "exogenous" data collection network sound promising—taking the lab outside to conduct distributed, crowdsourced data collection to address the authenticity of data. It’s a clever concept because even the most sophisticated lab cannot simulate the complex mechanical interactions of grabbing a live fish at a wet market. However, reality is harsh. Tactile data lacks the relatively unified representation of vision. Aligning over a dozen modalities—pressure, deformation, texture, stiffness, etc.—with vision and motor commands at millisecond-level spatiotemporal precision is an engineering challenge as daunting as embroidering in a storm. The so-called "world’s largest tactile dataset" and "open-source 10,000 hours" are more like planting a flag in a desert to prove water exists—yet an oasis remains far off.

In Daimon’s narrative, "defining standards" is a recurring theme. The release of the Daimon-Infinity dataset and the RobOmni Benchmark makes the intention clear: they don’t want to be just a technology provider but a rule-maker. In a chaotic period without industry standards, whoever first establishes a widely accepted evaluation system secures a pivotal ecosystem position. It’s a smart move but also risky. Standards can’t be established by a single company publishing a few papers; they require rigorous testing across real, diverse, and extreme scenarios and need recognition from both competitors and downstream customers. Otherwise, the "standard" becomes self-talk.

From a capital market perspective, the joint investment from Inovance Technology and China Telecom reflects industry consensus on the "physical interaction" track. Inovance, rooted in industrial automation, understands the value of "tactile sense" for precision manufacturing; China Telecom, betting on cloud-network integration and robotics, sees the explosive potential of "tactile intelligence" in service, inspection, and other scenarios. This funding isn’t just a bet on Daimon alone—it’s a wager on the belief that the "tactile-first" technological path will ultimately prove viable.

Yet between vision and reality always lies a deep chasm called "engineering." Daimon claims its self-developed visuotactile sensors have the world’s first shipment volume—a remarkable milestone, indicating that hardware groundwork has begun to pay off. But from sensors to stable, low-cost, easy-to-deploy modules, and then to enabling robots to reliably perform delicate tasks like "threading grapes" in dynamic home environments, countless pitfalls remain. A robot that can pour water perfectly may require tens of thousands of failures and hundreds of millions of data iterations. The flywheel has started turning, but to generate sufficient commercial momentum, it still needs time, patience, and more real investment.

Ultimately, Daimon’s story is a microcosm of embodied intelligence stepping from sci-fi imagination into reality. When the industry is no longer satisfied with somersaults and dances but starts grappling with how to "pick up a sponge just right," it signals that the true moment of application is near. Daimon has chosen a path that seems laborious but may be the most direct—letting robots re-learn humanity’s most primitive cognitive approach. The journey is long, but the direction may be right. After all, humans evolved over millions of years to learn tool use. Why should we expect robots to master it in a few years with just a few vision models? Touch may be the key to unlocking the real physical world. As for whether Daimon can ultimately "define the standards" and become the forger of this key, let’s calmly observe and give it time.

亿元融资砸向一家名不见经传的机器人公司,赌的是“手感”这个最基础、也最性感的人类本能。戴盟机器人拿到钱了,但更值得琢磨的是,它戳中了具身智能行业集体焦虑的那个点:所有炫酷的demo背后,机器人终究要和这个黏糊糊、硬邦邦、滑溜溜的物理世界发生真实接触。目前的行业逻辑,正从“跑得稳”、“看得见”的炫技阶段,被迫切换到“摸得准”、“握得住”的求生阶段。这场切换,不是升级,而是补课。

纯视觉路线造出的机器人,像一个高度近视又触觉失调的巨婴。它能看见鸡蛋,但不知道那是个易碎品;能识别海绵,但不理解压缩与回弹的力学关系。CEO段江哗的比喻很到位:视觉告诉你物体在哪里,触觉告诉你接触时发生了什么。这种“物理常识”的缺失,正是实验室机器人走进千家万户的最大鸿沟。戴盟的押注,本质上是在为机器人重新发明一种“婴儿期”的认知方式——通过触觉建立世界模型。这想法既复古又激进。

但建立触觉世界的门槛,远比搞一套视觉算法要高。这不仅仅是加个传感器的问题。戴盟宣称的“数据飞轮”和“外发式”采集网络,听起来很美好——把实验室搬出去,做分布式社会化采集,来解决数据真实性的问题。这确实是个巧思,因为再精密的实验室也模拟不了菜市场里手抓活鱼的复杂力学交互。然而,现实是残酷的。触觉数据没有视觉那样相对统一的表征方式,十几种模态(压力、形变、纹理、刚度…)要在毫秒级时空里与视觉、动作指令对齐,工程难度堪比在狂风暴雨中绣花。所谓“全球最大含触觉数据集”和“开源1万小时”,更像是在荒漠里率先插下的一面旗帜,证明这里有水,但离绿洲还远得很。

戴盟的叙事里,“定义标准”是个高频词。发布Daimon-Infinity数据集和RobOmni Benchmark,意图很明显:不想只做技术供应商,更想做规则制定者。在行业标准缺失的混沌期,谁先跑出能被广泛接受的评估体系,谁就占据了生态位。这步棋很聪明,但也很险。标准不是靠一家公司发几篇论文就能立住的,它需要经得起真实、多样、极端场景的反复淬炼,需要得到竞争对手和下游客户的共同认可。否则,这“标准”就成了自说自话。

从资本市场角度看,汇川和中国电信的联合投资,透露出产业方对“物理交互”赛道的共识。汇川做工业自动化出身,深知“手感”对精密制造的价值;中国电信押注云网与机器人,看得见“触觉智能”在服务、巡检等场景的爆发潜力。这笔钱,赌的不仅是戴盟一家,更是押注“触觉优先”这条技术路线终将跑通。

然而,愿景与现实之间,永远隔着一条名为“工程化”的深沟。公司自研视触觉传感器出货量全球第一,这是个了不起的里程碑,意味着硬件铺路已初见成效。但从传感器到稳定的、低成本的、易部署的模组,再到让机器人在动态家庭环境中可靠执行“串葡萄”这种精细操作,中间还有无数坑要填。一个能完美倒水的机器人,其背后可能需要数万次失败、数亿条数据迭代。飞轮已经开始转,但要形成足够的商业推力,还需要时间、耐心,以及更多真金白银的投入。

归根结底,戴盟的故事,是具身智能从科幻想象踏入凡尘的缩影。当行业不再满足于翻跟头和跳舞,而是开始纠结于如何“恰到好处地拿起一块海绵”时,说明真正的应用时刻临近了。戴盟选择了一条看似笨重却可能最直接的道路——让机器人重新学习人类最原始的认知方式。这条路很漫长,但方向或许没错。毕竟,人类进化了上亿年才学会使用工具,我们凭什么指望机器人靠几个视觉模型就能在几年内搞定?触觉,可能才是那把打开真实物理世界的钥匙。至于戴盟能否最终“定义标准”,成为这把钥匙的铸造者,我们不妨冷静旁观,让子弹再飞一会儿。

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

机器人 机器人 融资 融资 数据集 数据集
Share: 分享到: